diff --git a/llama_stack/providers/inline/tool_runtime/code_interpreter/code_execution.py b/llama_stack/providers/inline/tool_runtime/code_interpreter/code_execution.py index 6f4b25b9d..d7b2dbdef 100644 --- a/llama_stack/providers/inline/tool_runtime/code_interpreter/code_execution.py +++ b/llama_stack/providers/inline/tool_runtime/code_interpreter/code_execution.py @@ -76,6 +76,7 @@ class CodeExecutionRequest: only_last_cell_fail: bool = True seed: int = 0 strip_fpaths_in_stderr: bool = True + use_bwrap: bool = True class CodeExecutor: @@ -103,8 +104,6 @@ _set_seeds()\ script = "\n\n".join([seeds_prefix] + [CODE_ENV_PREFIX] + scripts) with tempfile.TemporaryDirectory() as dpath: - bwrap_prefix = "bwrap " + generate_bwrap_command(bind_dirs=[dpath]) - cmd = [*bwrap_prefix.split(), sys.executable, "-c", script] code_fpath = os.path.join(dpath, "code.py") with open(code_fpath, "w") as f: f.write(script) @@ -118,6 +117,13 @@ _set_seeds()\ MPLBACKEND="module://matplotlib_custom_backend", PYTHONPATH=f"{DIRNAME}:{python_path}", ) + + if req.use_bwrap: + bwrap_prefix = "bwrap " + generate_bwrap_command(bind_dirs=[dpath]) + cmd = [*bwrap_prefix.split(), sys.executable, "-c", script] + else: + cmd = [sys.executable, "-c", script] + stdout, stderr, returncode = do_subprocess( cmd=cmd, env=env, diff --git a/llama_stack/providers/inline/tool_runtime/code_interpreter/code_interpreter.py b/llama_stack/providers/inline/tool_runtime/code_interpreter/code_interpreter.py index 54f17f9a2..4b97914c5 100644 --- a/llama_stack/providers/inline/tool_runtime/code_interpreter/code_interpreter.py +++ b/llama_stack/providers/inline/tool_runtime/code_interpreter/code_interpreter.py @@ -6,6 +6,7 @@ import logging +import os import tempfile from typing import Any, Dict, List, Optional @@ -61,7 +62,9 @@ class CodeInterpreterToolRuntimeImpl(ToolsProtocolPrivate, ToolRuntime): async def invoke_tool(self, tool_name: str, kwargs: Dict[str, Any]) -> ToolInvocationResult: script = kwargs["code"] - req = CodeExecutionRequest(scripts=[script]) + # Use environment variable to control bwrap usage + force_disable_bwrap = os.environ.get("DISABLE_CODE_SANDBOX", "").lower() in ("1", "true", "yes") + req = CodeExecutionRequest(scripts=[script], use_bwrap=not force_disable_bwrap) res = self.code_executor.execute(req) pieces = [res["process_status"]] for out_type in ["stdout", "stderr"]: diff --git a/tests/integration/agents/test_agents.py b/tests/integration/agents/test_agents.py index a542e5403..f6bde8927 100644 --- a/tests/integration/agents/test_agents.py +++ b/tests/integration/agents/test_agents.py @@ -187,7 +187,7 @@ def test_builtin_tool_web_search(llama_stack_client_with_mocked_inference, agent messages=[ { "role": "user", - "content": "Search the web and tell me who the current CEO of Meta is.", + "content": "Search the web and tell me who the founder of Meta is.", } ], session_id=session_id, diff --git a/tests/integration/conftest.py b/tests/integration/conftest.py index f4fe9e8ff..bf1092c4a 100644 --- a/tests/integration/conftest.py +++ b/tests/integration/conftest.py @@ -6,12 +6,17 @@ import inspect import itertools import os +import platform import textwrap from dotenv import load_dotenv +from llama_stack.log import get_logger + from .report import Report +logger = get_logger(__name__, category="tests") + def pytest_configure(config): config.option.tbstyle = "short" @@ -24,6 +29,10 @@ def pytest_configure(config): key, value = env_var.split("=", 1) os.environ[key] = value + if platform.system() == "Darwin": # Darwin is the system name for macOS + os.environ["DISABLE_CODE_SANDBOX"] = "1" + logger.info("Setting DISABLE_CODE_SANDBOX=1 for macOS") + if config.getoption("--report"): config.pluginmanager.register(Report(config)) diff --git a/tests/integration/fixtures/recorded_responses/chat_completion.json b/tests/integration/fixtures/recorded_responses/chat_completion.json index 7234b6c31..30b7e0b4d 100644 --- a/tests/integration/fixtures/recorded_responses/chat_completion.json +++ b/tests/integration/fixtures/recorded_responses/chat_completion.json @@ -12535,7 +12535,23 @@ "value": "end_of_turn" } }, - "metrics": null + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 139 + }, + { + "metric": "completion_tokens", + "unit": null, + "value": 23 + }, + { + "metric": "total_tokens", + "unit": null, + "value": 162 + } + ] } } ], @@ -12589,7 +12605,7 @@ "data": { "event": { "delta": { - "text": "type\": \"function\", \"name\":", + "text": "type\": \"function\", \"name\": \"", "type": "text" }, "event_type": { @@ -12609,7 +12625,7 @@ "data": { "event": { "delta": { - "text": " \"get_boiling_point\", \"parameters\":", + "text": "get_boiling_point\", \"parameters\": {\"", "type": "text" }, "event_type": { @@ -12629,7 +12645,7 @@ "data": { "event": { "delta": { - "text": " {\"liquid_name\": \"polyjuice", + "text": "liquid_name\": \"polyjuice\", \"celcius\":", "type": "text" }, "event_type": { @@ -12649,7 +12665,7 @@ "data": { "event": { "delta": { - "text": "\", \"celcius\": \"false\"}}", + "text": " \"false\"}}", "type": "text" }, "event_type": { @@ -12679,7 +12695,7 @@ "celcius": "false", "liquid_name": "polyjuice" }, - "call_id": "bffe07d7-343f-49c4-bcff-d83c99fa7d4a", + "call_id": "fc7e2525-3e7b-47ff-8731-12dd7655dfd6", "tool_name": "get_boiling_point" }, "type": "tool_call" @@ -12720,7 +12736,23 @@ "value": "end_of_turn" } }, - "metrics": null + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 91 + }, + { + "metric": "completion_tokens", + "unit": null, + "value": 45 + }, + { + "metric": "total_tokens", + "unit": null, + "value": 136 + } + ] } } ], @@ -12774,7 +12806,7 @@ "data": { "event": { "delta": { - "text": " \"type\": \"function\",\n \"name\": \"get", + "text": " \"type\": \"function\",\n \"", "type": "text" }, "event_type": { @@ -12794,7 +12826,7 @@ "data": { "event": { "delta": { - "text": "_boiling_point\",\n \"parameters\": {\n \"liquid_name", + "text": "name\": \"get_boiling_point\",\n \"parameters\":", "type": "text" }, "event_type": { @@ -12814,7 +12846,7 @@ "data": { "event": { "delta": { - "text": "\": \"polyjuice\",\n ", + "text": " {\n \"liquid_name\": \"polyjuice\",\n \"celci", "type": "text" }, "event_type": { @@ -12834,7 +12866,7 @@ "data": { "event": { "delta": { - "text": " \"celcius\": \"true\"\n }\n}", + "text": "us\": \"true\"\n }\n}", "type": "text" }, "event_type": { @@ -12864,7 +12896,7 @@ "celcius": "true", "liquid_name": "polyjuice" }, - "call_id": "41ce6bfb-81c1-438d-8520-329c4446f1bc", + "call_id": "73212def-09c0-4a29-845e-149afb38fcd1", "tool_name": "get_boiling_point" }, "type": "tool_call" @@ -12905,7 +12937,23 @@ "value": "end_of_turn" } }, - "metrics": null + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 43 + }, + { + "metric": "completion_tokens", + "unit": null, + "value": 55 + }, + { + "metric": "total_tokens", + "unit": null, + "value": 98 + } + ] } } ], @@ -12959,7 +13007,27 @@ "data": { "event": { "delta": { - "text": " boiling point of polyjuice is -100\u00b0C.", + "text": " boiling point of polyjuice is", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " -100\u00b0C.", "type": "text" }, "event_type": { @@ -12994,7 +13062,23 @@ "value": "end_of_turn" } }, - "metrics": null + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 85 + }, + { + "metric": "completion_tokens", + "unit": null, + "value": 22 + }, + { + "metric": "total_tokens", + "unit": null, + "value": 107 + } + ] } } ], @@ -13048,7 +13132,27 @@ "data": { "event": { "delta": { - "text": " boiling point of polyjuice is -100\u00b0C.", + "text": " boiling point of polyjuice is -100", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "\u00b0C.", "type": "text" }, "event_type": { @@ -13083,13 +13187,230 @@ "value": "end_of_turn" } }, - "metrics": null + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 87 + }, + { + "metric": "completion_tokens", + "unit": null, + "value": 22 + }, + { + "metric": "total_tokens", + "unit": null, + "value": 109 + } + ] } } ], "type": "generator" }, "[[\"meta-llama/Llama-3.1-8B-Instruct\", [{\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"SystemMessage\", \"data\": {\"content\": \"You are a helpful assistant\", \"role\": \"system\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"UserMessage\", \"data\": {\"content\": \"Call get_boiling_point and answer What is the boiling point of polyjuice?\", \"context\": null, \"role\": \"user\"}}]], {\"response_format\": null, \"sampling_params\": {\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"SamplingParams\", \"data\": {\"max_tokens\": 0, \"repetition_penalty\": 1.0, \"strategy\": {\"temperature\": 0.0001, \"top_p\": 0.9, \"type\": \"top_p\"}}}, \"stream\": true, \"tool_config\": {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolConfig\", \"data\": {\"system_message_behavior\": {\"__enum__\": \"SystemMessageBehavior\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"append\"}, \"tool_choice\": {\"__enum__\": \"ToolChoice\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"auto\"}, \"tool_prompt_format\": null}}, \"tool_prompt_format\": null, \"tools\": [{\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"ToolDefinition\", \"data\": {\"description\": \"Returns the boiling point of a liquid in Celcius or Fahrenheit\", \"parameters\": {\"celcius\": {\"default\": true, \"description\": \"Whether to return the boiling point in Celcius\", \"param_type\": \"bool\", \"required\": false}, \"liquid_name\": {\"default\": null, \"description\": \"The name of the liquid\", \"param_type\": \"string\", \"required\": true}}, \"tool_name\": \"get_boiling_point\"}}]}]": { + "chunks": [ + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "start" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "parse_status": { + "__enum__": "ToolCallParseStatus", + "__module__": "llama_stack.apis.common.content_types", + "value": "started" + }, + "tool_call": "", + "type": "tool_call" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "parse_status": { + "__enum__": "ToolCallParseStatus", + "__module__": "llama_stack.apis.common.content_types", + "value": "in_progress" + }, + "tool_call": "{\"type\": \"function\", \"name", + "type": "tool_call" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "parse_status": { + "__enum__": "ToolCallParseStatus", + "__module__": "llama_stack.apis.common.content_types", + "value": "in_progress" + }, + "tool_call": "\": \"get_boiling_point\", \"parameters\": {\"liquid_name\": \"poly", + "type": "tool_call" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "parse_status": { + "__enum__": "ToolCallParseStatus", + "__module__": "llama_stack.apis.common.content_types", + "value": "in_progress" + }, + "tool_call": "juice\", \"celcius\": \"true\"}}", + "type": "tool_call" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "parse_status": { + "__enum__": "ToolCallParseStatus", + "__module__": "llama_stack.apis.common.content_types", + "value": "succeeded" + }, + "tool_call": { + "arguments": { + "celcius": "true", + "liquid_name": "polyjuice" + }, + "call_id": "e4b0121a-7b75-4e89-be40-d13021a3bb11", + "tool_name": "get_boiling_point" + }, + "type": "tool_call" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": { + "__enum__": "StopReason", + "__module__": "llama_stack.models.llama.datatypes", + "value": "end_of_turn" + } + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "complete" + }, + "logprobs": null, + "stop_reason": { + "__enum__": "StopReason", + "__module__": "llama_stack.models.llama.datatypes", + "value": "end_of_turn" + } + }, + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 37 + }, + { + "metric": "completion_tokens", + "unit": null, + "value": 10 + }, + { + "metric": "total_tokens", + "unit": null, + "value": 47 + } + ] + } + } + ], + "type": "generator" + }, + "[[\"meta-llama/Llama-3.1-8B-Instruct\", [{\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"SystemMessage\", \"data\": {\"content\": \"You are a helpful assistant\", \"role\": \"system\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"UserMessage\", \"data\": {\"content\": \"Call get_boiling_point and answer What is the boiling point of polyjuice?\", \"context\": null, \"role\": \"user\"}}]], {\"response_format\": null, \"sampling_params\": {\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"SamplingParams\", \"data\": {\"max_tokens\": 0, \"repetition_penalty\": 1.0, \"strategy\": {\"temperature\": 0.0001, \"top_p\": 0.9, \"type\": \"top_p\"}}}, \"stream\": true, \"tool_config\": {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolConfig\", \"data\": {\"system_message_behavior\": {\"__enum__\": \"SystemMessageBehavior\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"append\"}, \"tool_choice\": {\"__enum__\": \"ToolChoice\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"auto\"}, \"tool_prompt_format\": null}}, \"tool_prompt_format\": null, \"tools\": [{\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"ToolDefinition\", \"data\": {\"description\": \"Returns the boiling point of a liquid in Celcius or Fahrenheit\", \"parameters\": {\"celcius\": {\"default\": true, \"description\": \"Whether to return the boiling point in Celcius\", \"param_type\": \"bool\", \"required\": false}, \"liquid_name\": {\"default\": null, \"description\": \"The name of the liquid\", \"param_type\": \"string\", \"required\": true}}, \"tool_name\": \"get_boiling_point_with_metadata\"}}]}]": { "chunks": [ { "__module__": "llama_stack.apis.inference.inference", @@ -13172,7 +13493,7 @@ "__module__": "llama_stack.apis.common.content_types", "value": "in_progress" }, - "tool_call": "\", \"parameters\": {\"liquid_name\": \"polyjuice\", \"cel", + "tool_call": "_with_metadata\", \"parameters\": {\"liquid_name\": \"polyjuice\",", "type": "tool_call" }, "event_type": { @@ -13197,7 +13518,7 @@ "__module__": "llama_stack.apis.common.content_types", "value": "in_progress" }, - "tool_call": "cius\": \"true\"}}", + "tool_call": " \"celcius\": \"true\"}}", "type": "tool_call" }, "event_type": { @@ -13227,192 +13548,7 @@ "celcius": "true", "liquid_name": "polyjuice" }, - "call_id": "6161b956-9b68-4e88-87bf-e26a07d4c7ca", - "tool_name": "get_boiling_point" - }, - "type": "tool_call" - }, - "event_type": { - "__enum__": "ChatCompletionResponseEventType", - "__module__": "llama_stack.apis.inference.inference", - "value": "progress" - }, - "logprobs": null, - "stop_reason": { - "__enum__": "StopReason", - "__module__": "llama_stack.models.llama.datatypes", - "value": "end_of_turn" - } - }, - "metrics": null - } - }, - { - "__module__": "llama_stack.apis.inference.inference", - "__pydantic__": "ChatCompletionResponseStreamChunk", - "data": { - "event": { - "delta": { - "text": "", - "type": "text" - }, - "event_type": { - "__enum__": "ChatCompletionResponseEventType", - "__module__": "llama_stack.apis.inference.inference", - "value": "complete" - }, - "logprobs": null, - "stop_reason": { - "__enum__": "StopReason", - "__module__": "llama_stack.models.llama.datatypes", - "value": "end_of_turn" - } - }, - "metrics": null - } - } - ], - "type": "generator" - }, - "[[\"meta-llama/Llama-3.1-8B-Instruct\", [{\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"SystemMessage\", \"data\": {\"content\": \"You are a helpful assistant\", \"role\": \"system\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"UserMessage\", \"data\": {\"content\": \"Call get_boiling_point and answer What is the boiling point of polyjuice?\", \"context\": null, \"role\": \"user\"}}]], {\"response_format\": null, \"sampling_params\": {\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"SamplingParams\", \"data\": {\"max_tokens\": 0, \"repetition_penalty\": 1.0, \"strategy\": {\"temperature\": 0.0001, \"top_p\": 0.9, \"type\": \"top_p\"}}}, \"stream\": true, \"tool_config\": {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolConfig\", \"data\": {\"system_message_behavior\": {\"__enum__\": \"SystemMessageBehavior\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"append\"}, \"tool_choice\": {\"__enum__\": \"ToolChoice\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"auto\"}, \"tool_prompt_format\": null}}, \"tool_prompt_format\": null, \"tools\": [{\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"ToolDefinition\", \"data\": {\"description\": \"Returns the boiling point of a liquid in Celcius or Fahrenheit\", \"parameters\": {\"celcius\": {\"default\": true, \"description\": \"Whether to return the boiling point in Celcius\", \"param_type\": \"bool\", \"required\": false}, \"liquid_name\": {\"default\": null, \"description\": \"The name of the liquid\", \"param_type\": \"string\", \"required\": true}}, \"tool_name\": \"get_boiling_point_with_metadata\"}}]}]": { - "chunks": [ - { - "__module__": "llama_stack.apis.inference.inference", - "__pydantic__": "ChatCompletionResponseStreamChunk", - "data": { - "event": { - "delta": { - "text": "", - "type": "text" - }, - "event_type": { - "__enum__": "ChatCompletionResponseEventType", - "__module__": "llama_stack.apis.inference.inference", - "value": "start" - }, - "logprobs": null, - "stop_reason": null - }, - "metrics": null - } - }, - { - "__module__": "llama_stack.apis.inference.inference", - "__pydantic__": "ChatCompletionResponseStreamChunk", - "data": { - "event": { - "delta": { - "parse_status": { - "__enum__": "ToolCallParseStatus", - "__module__": "llama_stack.apis.common.content_types", - "value": "started" - }, - "tool_call": "", - "type": "tool_call" - }, - "event_type": { - "__enum__": "ChatCompletionResponseEventType", - "__module__": "llama_stack.apis.inference.inference", - "value": "progress" - }, - "logprobs": null, - "stop_reason": null - }, - "metrics": null - } - }, - { - "__module__": "llama_stack.apis.inference.inference", - "__pydantic__": "ChatCompletionResponseStreamChunk", - "data": { - "event": { - "delta": { - "parse_status": { - "__enum__": "ToolCallParseStatus", - "__module__": "llama_stack.apis.common.content_types", - "value": "in_progress" - }, - "tool_call": "{\"type\": \"function\", \"name\": \"get_boiling", - "type": "tool_call" - }, - "event_type": { - "__enum__": "ChatCompletionResponseEventType", - "__module__": "llama_stack.apis.inference.inference", - "value": "progress" - }, - "logprobs": null, - "stop_reason": null - }, - "metrics": null - } - }, - { - "__module__": "llama_stack.apis.inference.inference", - "__pydantic__": "ChatCompletionResponseStreamChunk", - "data": { - "event": { - "delta": { - "parse_status": { - "__enum__": "ToolCallParseStatus", - "__module__": "llama_stack.apis.common.content_types", - "value": "in_progress" - }, - "tool_call": "_point_with_metadata\", \"parameters\": {\"liquid_name\": \"polyju", - "type": "tool_call" - }, - "event_type": { - "__enum__": "ChatCompletionResponseEventType", - "__module__": "llama_stack.apis.inference.inference", - "value": "progress" - }, - "logprobs": null, - "stop_reason": null - }, - "metrics": null - } - }, - { - "__module__": "llama_stack.apis.inference.inference", - "__pydantic__": "ChatCompletionResponseStreamChunk", - "data": { - "event": { - "delta": { - "parse_status": { - "__enum__": "ToolCallParseStatus", - "__module__": "llama_stack.apis.common.content_types", - "value": "in_progress" - }, - "tool_call": "ice\", \"celcius\": \"true\"}}", - "type": "tool_call" - }, - "event_type": { - "__enum__": "ChatCompletionResponseEventType", - "__module__": "llama_stack.apis.inference.inference", - "value": "progress" - }, - "logprobs": null, - "stop_reason": null - }, - "metrics": null - } - }, - { - "__module__": "llama_stack.apis.inference.inference", - "__pydantic__": "ChatCompletionResponseStreamChunk", - "data": { - "event": { - "delta": { - "parse_status": { - "__enum__": "ToolCallParseStatus", - "__module__": "llama_stack.apis.common.content_types", - "value": "succeeded" - }, - "tool_call": { - "arguments": { - "celcius": "true", - "liquid_name": "polyjuice" - }, - "call_id": "11da4a37-d7ad-468a-98c8-0f1e295d14a9", + "call_id": "11c04896-2b7b-49bd-b832-47a1c9f3796f", "tool_name": "get_boiling_point_with_metadata" }, "type": "tool_call" @@ -13453,7 +13589,23 @@ "value": "end_of_turn" } }, - "metrics": null + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 37 + }, + { + "metric": "completion_tokens", + "unit": null, + "value": 10 + }, + { + "metric": "total_tokens", + "unit": null, + "value": 47 + } + ] } } ], @@ -13507,7 +13659,27 @@ "data": { "event": { "delta": { - "text": " customer smiled and said \"hello\" to the friendly store clerk.", + "text": " customer smiled and said \"hello\" to the", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " friendly store clerk.", "type": "text" }, "event_type": { @@ -13542,7 +13714,23 @@ "value": "end_of_turn" } }, - "metrics": null + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 30 + }, + { + "metric": "completion_tokens", + "unit": null, + "value": 24 + }, + { + "metric": "total_tokens", + "unit": null, + "value": 54 + } + ] } } ], @@ -13957,6 +14145,1165 @@ ], "type": "generator" }, + "[[\"meta-llama/Llama-3.1-8B-Instruct\", [{\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"SystemMessage\", \"data\": {\"content\": \"You are a helpful assistant\", \"role\": \"system\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"UserMessage\", \"data\": {\"content\": \"Here is a csv file, can you describe it?\", \"context\": null, \"role\": \"user\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolResponseMessage\", \"data\": {\"call_id\": \"\", \"content\": [{\"text\": \"# User provided a file accessible to you at \\\"\"\\nYou can use code_interpreter to load and inspect it.\", \"type\": \"text\"}], \"role\": \"tool\", \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"CompletionMessage\", \"data\": {\"content\": \"\", \"role\": \"assistant\", \"stop_reason\": {\"__enum__\": \"StopReason\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"end_of_turn\"}, \"tool_calls\": [{\"arguments\": {\"code\": \"import pandas as pd\\ndf = pd.read_csv(\\\"\")\\nprint(df.head())\"}, \"call_id\": \"\", \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}]}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolResponseMessage\", \"data\": {\"call_id\": \"\", \"content\": \"completed\\n[stdout]\\nYear Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec\\n0 2014 1.6 1.6 1.7 1.8 2.0 1.9 1.9 1.7 1.7 1.8 1.7 1.6\\n1 2015 1.6 1.7 1.8 1.8 1.7 1.8 1.8 1.8 1.9 1.9 2.0 2.1\\n2 2016 2.2 2.3 2.2 2.1 2.2 2.2 2.2 2.3 2.2 2.1 2.1 2.2\\n3 2017 2.3 2.2 2.0 1.9 1.7 1.7 1.7 1.7 1.7 1.8 1.7 1.8\\n4 2018 1.8 1.8 2.1 2.1 2.2 2.3 2.4 2.2 2.2 2.1 2.2 2.2\\n[/stdout]\", \"role\": \"tool\", \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}}]], {\"response_format\": null, \"sampling_params\": {\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"SamplingParams\", \"data\": {\"max_tokens\": 0, \"repetition_penalty\": 1.0, \"strategy\": {\"temperature\": 0.0001, \"top_p\": 0.9, \"type\": \"top_p\"}}}, \"stream\": true, \"tool_config\": {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolConfig\", \"data\": {\"system_message_behavior\": {\"__enum__\": \"SystemMessageBehavior\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"append\"}, \"tool_choice\": {\"__enum__\": \"ToolChoice\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"auto\"}, \"tool_prompt_format\": null}}, \"tool_prompt_format\": null, \"tools\": [{\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"ToolDefinition\", \"data\": {\"description\": \"Search for information in a database.\", \"parameters\": {\"query\": {\"default\": null, \"description\": \"The query to search for. Can be a natural language sentence or keywords.\", \"param_type\": \"string\", \"required\": true}}, \"tool_name\": \"knowledge_search\"}}, {\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"ToolDefinition\", \"data\": {\"description\": \"Execute code\", \"parameters\": {\"code\": {\"default\": null, \"description\": \"The code to execute\", \"param_type\": \"string\", \"required\": true}}, \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}}]}]": { + "chunks": [ + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "start" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "The", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " csv file contains data on inflation rates for each month of the year from", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " 2014 to 2018. The columns are:\n\n- Year", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": ": The year of the inflation rate\n-", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " Jan to Dec: The inflation rate for each month of", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " the year\n\nThe inflation rates are all in the range of ", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "1.6 to 2.4, indicating", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " a relatively stable inflation rate over the years.", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "complete" + }, + "logprobs": null, + "stop_reason": { + "__enum__": "StopReason", + "__module__": "llama_stack.models.llama.datatypes", + "value": "end_of_turn" + } + }, + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 471 + }, + { + "metric": "completion_tokens", + "unit": null, + "value": 91 + }, + { + "metric": "total_tokens", + "unit": null, + "value": 562 + } + ] + } + } + ], + "type": "generator" + }, + "[[\"meta-llama/Llama-3.1-8B-Instruct\", [{\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"SystemMessage\", \"data\": {\"content\": \"You are a helpful assistant\", \"role\": \"system\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"UserMessage\", \"data\": {\"content\": \"Here is a csv file, can you describe it?\", \"context\": null, \"role\": \"user\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolResponseMessage\", \"data\": {\"call_id\": \"\", \"content\": [{\"text\": \"# User provided a file accessible to you at \\\"\"\\nYou can use code_interpreter to load and inspect it.\", \"type\": \"text\"}], \"role\": \"tool\", \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"CompletionMessage\", \"data\": {\"content\": \"\", \"role\": \"assistant\", \"stop_reason\": {\"__enum__\": \"StopReason\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"end_of_turn\"}, \"tool_calls\": [{\"arguments\": {\"code\": \"import pandas as pd\\nimport code_interpreter\\n\\n# Load the CSV file\\ndf = pd.read_csv(\\\"\")\\n\\n# Print the first few rows of the dataframe\\nprint(df.head())\\n\\n# Print the data types of each column\\nprint(df.dtypes)\\n\\n# Print the summary statistics of the dataframe\\nprint(df.describe())\"}, \"call_id\": \"\", \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}]}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolResponseMessage\", \"data\": {\"call_id\": \"\", \"content\": \"completed\\n[stderr]\\nTraceback (most recent call last):\\n line 142, in \\n line 23, in \\n from .code_execution import CodeExecutionContext, CodeExecutionRequest, CodeExecutor\\nImportError: attempted relative import with no known parent package\\n[/stderr]\", \"role\": \"tool\", \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"CompletionMessage\", \"data\": {\"content\": \"\", \"role\": \"assistant\", \"stop_reason\": {\"__enum__\": \"StopReason\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"end_of_turn\"}, \"tool_calls\": [{\"arguments\": {\"code\": \"import pandas as pd\\nimport code_interpreter\\n\\n# Load the CSV file\\ndf = pd.read_csv(\\\"\")\\n\\n# Print the first few rows of the dataframe\\nprint(df.head())\\n\\n# Print the data types of each column\\nprint(df.dtypes)\\n\\n# Print the summary statistics of the dataframe\\nprint(df.describe())\"}, \"call_id\": \"\", \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}]}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolResponseMessage\", \"data\": {\"call_id\": \"\", \"content\": \"completed\\n[stderr]\\nTraceback (most recent call last):\\n line 142, in \\n line 23, in \\n from .code_execution import CodeExecutionContext, CodeExecutionRequest, CodeExecutor\\nImportError: attempted relative import with no known parent package\\n[/stderr]\", \"role\": \"tool\", \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}}]], {\"response_format\": null, \"sampling_params\": {\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"SamplingParams\", \"data\": {\"max_tokens\": 0, \"repetition_penalty\": 1.0, \"strategy\": {\"temperature\": 0.0001, \"top_p\": 0.9, \"type\": \"top_p\"}}}, \"stream\": true, \"tool_config\": {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolConfig\", \"data\": {\"system_message_behavior\": {\"__enum__\": \"SystemMessageBehavior\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"append\"}, \"tool_choice\": {\"__enum__\": \"ToolChoice\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"auto\"}, \"tool_prompt_format\": null}}, \"tool_prompt_format\": null, \"tools\": [{\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"ToolDefinition\", \"data\": {\"description\": \"Search for information in a database.\", \"parameters\": {\"query\": {\"default\": null, \"description\": \"The query to search for. Can be a natural language sentence or keywords.\", \"param_type\": \"string\", \"required\": true}}, \"tool_name\": \"knowledge_search\"}}, {\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"ToolDefinition\", \"data\": {\"description\": \"Execute code\", \"parameters\": {\"code\": {\"default\": null, \"description\": \"The code to execute\", \"param_type\": \"string\", \"required\": true}}, \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}}]}]": { + "chunks": [ + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "start" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "The", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " error message indicates that there is an issue with the import statement. However", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": ", the code provided does not contain any import statements that", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " would cause this error.\n\nTo provide a more accurate answer, I", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " would need to know the contents of the CSV file or more information", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " about the error message.\n\nHowever, based on the code provided, it", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " seems like the code is trying to load a CSV", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " file and print some basic information about it. If the file is", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " not found or there is an issue with the file", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " path, this could cause an error.\n\nHere is a revised version", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " of the code that includes some error handling:\n\n``", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "`\nimport pandas as pd\nimport code_interpreter\n\ntry:\n", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " # Load the CSV file\n df = pd.read_csv(\"/", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "var/folders/cz/vyh7y1d11x", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "g881lsxsshnc5c0000gn/T/tmp", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "_d_cdeif/6TpkUAo", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "0inflation.csv\")\n\n # Print the first few rows of the", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " dataframe\n print(df.head())\n\n # Print the", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " data types of each column\n print(df", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": ".dtypes)\n\n # Print the summary statistics of the dataframe\n", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " print(df.describe())\n\nexcept FileNotFoundError:\n print(\"The file was", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " not found.\")\nexcept pd.errors.EmptyDataError:\n print(\"The", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " file is empty.\")\nexcept pd.errors.ParserError:\n print(\"An", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " error occurred while parsing the file.\")\nexcept Exception as e:\n print", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "(\"An error occurred: \", str(e))\n```\n\nThis code will", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " catch specific exceptions that could occur when loading the CSV file and print a", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " more informative error message.", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "complete" + }, + "logprobs": null, + "stop_reason": { + "__enum__": "StopReason", + "__module__": "llama_stack.models.llama.datatypes", + "value": "end_of_turn" + } + }, + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 391 + }, + { + "metric": "completion_tokens", + "unit": null, + "value": 330 + }, + { + "metric": "total_tokens", + "unit": null, + "value": 721 + } + ] + } + } + ], + "type": "generator" + }, + "[[\"meta-llama/Llama-3.1-8B-Instruct\", [{\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"SystemMessage\", \"data\": {\"content\": \"You are a helpful assistant\", \"role\": \"system\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"UserMessage\", \"data\": {\"content\": \"Here is a csv file, can you describe it?\", \"context\": null, \"role\": \"user\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolResponseMessage\", \"data\": {\"call_id\": \"\", \"content\": [{\"text\": \"# User provided a file accessible to you at \\\"\"\\nYou can use code_interpreter to load and inspect it.\", \"type\": \"text\"}], \"role\": \"tool\", \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"CompletionMessage\", \"data\": {\"content\": \"\", \"role\": \"assistant\", \"stop_reason\": {\"__enum__\": \"StopReason\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"end_of_turn\"}, \"tool_calls\": [{\"arguments\": {\"code\": \"import pandas as pd\\nimport code_interpreter\\n\\n# Load the CSV file\\ndf = pd.read_csv(\\\"\")\\n\\n# Print the first few rows of the dataframe\\nprint(df.head())\\n\\n# Print the data types of each column\\nprint(df.dtypes)\\n\\n# Print the summary statistics of the dataframe\\nprint(df.describe())\"}, \"call_id\": \"\", \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}]}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolResponseMessage\", \"data\": {\"call_id\": \"\", \"content\": \"completed\\n[stderr]\\nTraceback (most recent call last):\\n line 142, in \\n line 23, in \\n from .code_execution import CodeExecutionContext, CodeExecutionRequest, CodeExecutor\\nImportError: attempted relative import with no known parent package\\n[/stderr]\", \"role\": \"tool\", \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}}]], {\"response_format\": null, \"sampling_params\": {\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"SamplingParams\", \"data\": {\"max_tokens\": 0, \"repetition_penalty\": 1.0, \"strategy\": {\"temperature\": 0.0001, \"top_p\": 0.9, \"type\": \"top_p\"}}}, \"stream\": true, \"tool_config\": {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolConfig\", \"data\": {\"system_message_behavior\": {\"__enum__\": \"SystemMessageBehavior\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"append\"}, \"tool_choice\": {\"__enum__\": \"ToolChoice\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"auto\"}, \"tool_prompt_format\": null}}, \"tool_prompt_format\": null, \"tools\": [{\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"ToolDefinition\", \"data\": {\"description\": \"Search for information in a database.\", \"parameters\": {\"query\": {\"default\": null, \"description\": \"The query to search for. Can be a natural language sentence or keywords.\", \"param_type\": \"string\", \"required\": true}}, \"tool_name\": \"knowledge_search\"}}, {\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"ToolDefinition\", \"data\": {\"description\": \"Execute code\", \"parameters\": {\"code\": {\"default\": null, \"description\": \"The code to execute\", \"param_type\": \"string\", \"required\": true}}, \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}}]}]": { + "chunks": [ + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "start" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "parse_status": { + "__enum__": "ToolCallParseStatus", + "__module__": "llama_stack.apis.common.content_types", + "value": "started" + }, + "tool_call": "", + "type": "tool_call" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "parse_status": { + "__enum__": "ToolCallParseStatus", + "__module__": "llama_stack.apis.common.content_types", + "value": "in_progress" + }, + "tool_call": "import pandas as pd\nimport code_interpreter\n\n", + "type": "tool_call" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "parse_status": { + "__enum__": "ToolCallParseStatus", + "__module__": "llama_stack.apis.common.content_types", + "value": "in_progress" + }, + "tool_call": "# Load the CSV file\ndf = pd.read_csv(\"/var", + "type": "tool_call" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "parse_status": { + "__enum__": "ToolCallParseStatus", + "__module__": "llama_stack.apis.common.content_types", + "value": "in_progress" + }, + "tool_call": "/folders/cz/vyh7y1d11xg881", + "type": "tool_call" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "parse_status": { + "__enum__": "ToolCallParseStatus", + "__module__": "llama_stack.apis.common.content_types", + "value": "in_progress" + }, + "tool_call": "lsxsshnc5c0000gn/T/tmp_d_cdeif", + "type": "tool_call" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "parse_status": { + "__enum__": "ToolCallParseStatus", + "__module__": "llama_stack.apis.common.content_types", + "value": "in_progress" + }, + "tool_call": "/6TpkUAo0inflation.csv\")\n\n# Print the", + "type": "tool_call" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "parse_status": { + "__enum__": "ToolCallParseStatus", + "__module__": "llama_stack.apis.common.content_types", + "value": "in_progress" + }, + "tool_call": " first few rows of the dataframe\nprint(df.head())\n\n# Print the", + "type": "tool_call" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "parse_status": { + "__enum__": "ToolCallParseStatus", + "__module__": "llama_stack.apis.common.content_types", + "value": "in_progress" + }, + "tool_call": " data types of each column\nprint(df.dtypes)\n\n# Print the", + "type": "tool_call" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "parse_status": { + "__enum__": "ToolCallParseStatus", + "__module__": "llama_stack.apis.common.content_types", + "value": "in_progress" + }, + "tool_call": " summary statistics of the dataframe\nprint(df.describe())", + "type": "tool_call" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "parse_status": { + "__enum__": "ToolCallParseStatus", + "__module__": "llama_stack.apis.common.content_types", + "value": "succeeded" + }, + "tool_call": { + "arguments": { + "code": "import pandas as pd\nimport code_interpreter\n\n# Load the CSV file\ndf = pd.read_csv(\"/var/folders/cz/vyh7y1d11xg881lsxsshnc5c0000gn/T/tmp_d_cdeif/6TpkUAo0inflation.csv\")\n\n# Print the first few rows of the dataframe\nprint(df.head())\n\n# Print the data types of each column\nprint(df.dtypes)\n\n# Print the summary statistics of the dataframe\nprint(df.describe())" + }, + "call_id": "fa1b393f-3fc7-416f-98ab-05d879def880", + "tool_name": { + "__enum__": "BuiltinTool", + "__module__": "llama_stack.models.llama.datatypes", + "value": "code_interpreter" + } + }, + "type": "tool_call" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": { + "__enum__": "StopReason", + "__module__": "llama_stack.models.llama.datatypes", + "value": "end_of_turn" + } + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "complete" + }, + "logprobs": null, + "stop_reason": { + "__enum__": "StopReason", + "__module__": "llama_stack.models.llama.datatypes", + "value": "end_of_turn" + } + }, + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 214 + }, + { + "metric": "completion_tokens", + "unit": null, + "value": 10 + }, + { + "metric": "total_tokens", + "unit": null, + "value": 224 + } + ] + } + } + ], + "type": "generator" + }, "[[\"meta-llama/Llama-3.1-8B-Instruct\", [{\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"SystemMessage\", \"data\": {\"content\": \"You are a helpful assistant\", \"role\": \"system\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"UserMessage\", \"data\": {\"content\": \"Here is a csv file, can you describe it?\", \"context\": null, \"role\": \"user\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolResponseMessage\", \"data\": {\"call_id\": \"\", \"content\": [{\"text\": \"# User provided a file accessible to you at \\\"\"\\nYou can use code_interpreter to load and inspect it.\", \"type\": \"text\"}], \"role\": \"tool\", \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"CompletionMessage\", \"data\": {\"content\": \"\", \"role\": \"assistant\", \"stop_reason\": {\"__enum__\": \"StopReason\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"end_of_turn\"}, \"tool_calls\": [{\"arguments\": {\"code\": \"import pandas as pd\\nimport code_interpreter\\n\\n# Load the CSV file\\ndf = pd.read_csv(\\\"\")\\n\\n# Print the first few rows of the dataframe\\nprint(df.head())\\n\\n# Print the data types of each column\\nprint(df.dtypes)\\n\\n# Print the summary statistics of the dataframe\\nprint(df.describe())\"}, \"call_id\": \"\", \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}]}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolResponseMessage\", \"data\": {\"call_id\": \"\", \"content\": \"completed\\n[stderr]\\nTraceback (most recent call last):\\n line 5, in \\n from bwrap.core import main\\nModuleNotFoundError: No module named 'bwrap.core'\\n[/stderr]\", \"role\": \"tool\", \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"CompletionMessage\", \"data\": {\"content\": \"\", \"role\": \"assistant\", \"stop_reason\": {\"__enum__\": \"StopReason\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"end_of_turn\"}, \"tool_calls\": [{\"arguments\": {\"code\": \"import pandas as pd\\nimport code_interpreter\\n\\n# Load the CSV file\\ndf = pd.read_csv(\\\"\")\\n\\n# Print the first few rows of the dataframe\\nprint(df.head())\\n\\n# Print the data types of each column\\nprint(df.dtypes)\\n\\n# Print the summary statistics of the dataframe\\nprint(df.describe())\"}, \"call_id\": \"\", \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}]}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolResponseMessage\", \"data\": {\"call_id\": \"\", \"content\": \"completed\\n[stderr]\\nTraceback (most recent call last):\\n line 5, in \\n from bwrap.core import main\\nModuleNotFoundError: No module named 'bwrap.core'\\n[/stderr]\", \"role\": \"tool\", \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}}]], {\"response_format\": null, \"sampling_params\": {\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"SamplingParams\", \"data\": {\"max_tokens\": 0, \"repetition_penalty\": 1.0, \"strategy\": {\"temperature\": 0.0001, \"top_p\": 0.9, \"type\": \"top_p\"}}}, \"stream\": true, \"tool_config\": {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolConfig\", \"data\": {\"system_message_behavior\": {\"__enum__\": \"SystemMessageBehavior\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"append\"}, \"tool_choice\": {\"__enum__\": \"ToolChoice\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"auto\"}, \"tool_prompt_format\": null}}, \"tool_prompt_format\": null, \"tools\": [{\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"ToolDefinition\", \"data\": {\"description\": \"Search for information in a database.\", \"parameters\": {\"query\": {\"default\": null, \"description\": \"The query to search for. Can be a natural language sentence or keywords.\", \"param_type\": \"string\", \"required\": true}}, \"tool_name\": \"knowledge_search\"}}, {\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"ToolDefinition\", \"data\": {\"description\": \"Execute code\", \"parameters\": {\"code\": {\"default\": null, \"description\": \"The code to execute\", \"param_type\": \"string\", \"required\": true}}, \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}}]}]": { "chunks": [ { @@ -15046,7 +16393,7 @@ "__module__": "llama_stack.apis.common.content_types", "value": "in_progress" }, - "tool_call": "import pandas as pd\n", + "tool_call": "import pandas as pd\ndf = pd.read_csv(\"/var/folders/c", "type": "tool_call" }, "event_type": { @@ -15071,7 +16418,7 @@ "__module__": "llama_stack.apis.common.content_types", "value": "in_progress" }, - "tool_call": "import code_interpreter\n\n# Load the CSV file\ndf =", + "tool_call": "z/vyh7y1d11xg881lsxsshnc5", "type": "tool_call" }, "event_type": { @@ -15096,7 +16443,7 @@ "__module__": "llama_stack.apis.common.content_types", "value": "in_progress" }, - "tool_call": " pd.read_csv(\"/var/folders", + "tool_call": "c0000gn/T/tmpe8u6r9sz/R", "type": "tool_call" }, "event_type": { @@ -15121,157 +16468,7 @@ "__module__": "llama_stack.apis.common.content_types", "value": "in_progress" }, - "tool_call": "/cz/vyh7y1d11xg881", - "type": "tool_call" - }, - "event_type": { - "__enum__": "ChatCompletionResponseEventType", - "__module__": "llama_stack.apis.inference.inference", - "value": "progress" - }, - "logprobs": null, - "stop_reason": null - }, - "metrics": null - } - }, - { - "__module__": "llama_stack.apis.inference.inference", - "__pydantic__": "ChatCompletionResponseStreamChunk", - "data": { - "event": { - "delta": { - "parse_status": { - "__enum__": "ToolCallParseStatus", - "__module__": "llama_stack.apis.common.content_types", - "value": "in_progress" - }, - "tool_call": "lsxsshnc5c0000gn/T/tmp4ed7", - "type": "tool_call" - }, - "event_type": { - "__enum__": "ChatCompletionResponseEventType", - "__module__": "llama_stack.apis.inference.inference", - "value": "progress" - }, - "logprobs": null, - "stop_reason": null - }, - "metrics": null - } - }, - { - "__module__": "llama_stack.apis.inference.inference", - "__pydantic__": "ChatCompletionResponseStreamChunk", - "data": { - "event": { - "delta": { - "parse_status": { - "__enum__": "ToolCallParseStatus", - "__module__": "llama_stack.apis.common.content_types", - "value": "in_progress" - }, - "tool_call": "p2bg/Csr659svinflation.csv\")\n\n# Print", - "type": "tool_call" - }, - "event_type": { - "__enum__": "ChatCompletionResponseEventType", - "__module__": "llama_stack.apis.inference.inference", - "value": "progress" - }, - "logprobs": null, - "stop_reason": null - }, - "metrics": null - } - }, - { - "__module__": "llama_stack.apis.inference.inference", - "__pydantic__": "ChatCompletionResponseStreamChunk", - "data": { - "event": { - "delta": { - "parse_status": { - "__enum__": "ToolCallParseStatus", - "__module__": "llama_stack.apis.common.content_types", - "value": "in_progress" - }, - "tool_call": " the first few rows of the dataframe\nprint(df.head())\n\n#", - "type": "tool_call" - }, - "event_type": { - "__enum__": "ChatCompletionResponseEventType", - "__module__": "llama_stack.apis.inference.inference", - "value": "progress" - }, - "logprobs": null, - "stop_reason": null - }, - "metrics": null - } - }, - { - "__module__": "llama_stack.apis.inference.inference", - "__pydantic__": "ChatCompletionResponseStreamChunk", - "data": { - "event": { - "delta": { - "parse_status": { - "__enum__": "ToolCallParseStatus", - "__module__": "llama_stack.apis.common.content_types", - "value": "in_progress" - }, - "tool_call": " Print the data types of each column\nprint(df.dtypes)\n\n", - "type": "tool_call" - }, - "event_type": { - "__enum__": "ChatCompletionResponseEventType", - "__module__": "llama_stack.apis.inference.inference", - "value": "progress" - }, - "logprobs": null, - "stop_reason": null - }, - "metrics": null - } - }, - { - "__module__": "llama_stack.apis.inference.inference", - "__pydantic__": "ChatCompletionResponseStreamChunk", - "data": { - "event": { - "delta": { - "parse_status": { - "__enum__": "ToolCallParseStatus", - "__module__": "llama_stack.apis.common.content_types", - "value": "in_progress" - }, - "tool_call": "# Print the summary statistics of the dataframe", - "type": "tool_call" - }, - "event_type": { - "__enum__": "ChatCompletionResponseEventType", - "__module__": "llama_stack.apis.inference.inference", - "value": "progress" - }, - "logprobs": null, - "stop_reason": null - }, - "metrics": null - } - }, - { - "__module__": "llama_stack.apis.inference.inference", - "__pydantic__": "ChatCompletionResponseStreamChunk", - "data": { - "event": { - "delta": { - "parse_status": { - "__enum__": "ToolCallParseStatus", - "__module__": "llama_stack.apis.common.content_types", - "value": "in_progress" - }, - "tool_call": "\nprint(df.describe())", + "tool_call": "ChoI8s0inflation.csv\")\nprint(df.head())", "type": "tool_call" }, "event_type": { @@ -15298,9 +16495,9 @@ }, "tool_call": { "arguments": { - "code": "import pandas as pd\nimport code_interpreter\n\n# Load the CSV file\ndf = pd.read_csv(\"/var/folders/cz/vyh7y1d11xg881lsxsshnc5c0000gn/T/tmp4ed7p2bg/Csr659svinflation.csv\")\n\n# Print the first few rows of the dataframe\nprint(df.head())\n\n# Print the data types of each column\nprint(df.dtypes)\n\n# Print the summary statistics of the dataframe\nprint(df.describe())" + "code": "import pandas as pd\ndf = pd.read_csv(\"/var/folders/cz/vyh7y1d11xg881lsxsshnc5c0000gn/T/tmpe8u6r9sz/RChoI8s0inflation.csv\")\nprint(df.head())" }, - "call_id": "8aeab20b-341b-4349-84dc-3e3c3299d713", + "call_id": "35e85870-f8f3-44f4-8879-e7b02a2805f6", "tool_name": { "__enum__": "BuiltinTool", "__module__": "llama_stack.models.llama.datatypes", @@ -15345,7 +16542,23 @@ "value": "end_of_turn" } }, - "metrics": null + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 37 + }, + { + "metric": "completion_tokens", + "unit": null, + "value": 10 + }, + { + "metric": "total_tokens", + "unit": null, + "value": 47 + } + ] } } ], @@ -17611,6 +18824,670 @@ ], "type": "generator" }, + "[[\"meta-llama/Llama-3.1-8B-Instruct\", [{\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"SystemMessage\", \"data\": {\"content\": \"You are a helpful assistant\", \"role\": \"system\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"UserMessage\", \"data\": {\"content\": \"Here is a csv, can you describe it?\", \"context\": null, \"role\": \"user\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"CompletionMessage\", \"data\": {\"content\": \"\", \"role\": \"assistant\", \"stop_reason\": {\"__enum__\": \"StopReason\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"end_of_turn\"}, \"tool_calls\": [{\"arguments\": {\"code\": \"import pandas as pd\\n# Load data\\ndf = pd.read_csv(\\\"\")\\n# Rows\\nprint(\\\"Number of rows and columns in the data:\\\", df.shape)\\n# Columns\\nprint(\\\"Columns of the data are:\\\", len(df.columns))\\n# Column names\\nprint(\\\"Columns of the data are:\\\", df.columns)\\n# Column dtypes\\nprint(\\\"Datatype of the columns are:\\\", df.dtypes)\"}, \"call_id\": \"\", \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}]}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolResponseMessage\", \"data\": {\"call_id\": \"\", \"content\": \"completed\\n[stdout]\\nNumber of rows and columns in the data: (10, 13)\\nColumns of the data are: 13\\nColumns of the data are: Index(['Year', 'Jan', 'Feb', 'Mar', 'Apr', 'May', 'Jun', 'Jul', 'Aug', 'Sep',\\n 'Oct', 'Nov', 'Dec'],\\n dtype='object')\\nDatatype of the columns are: Year int64\\nJan float64\\nFeb float64\\nMar float64\\nApr float64\\nMay float64\\nJun float64\\nJul float64\\nAug float64\\nSep float64\\nOct float64\\nNov float64\\nDec float64\\ndtype: object\\n[/stdout]\", \"role\": \"tool\", \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"CompletionMessage\", \"data\": {\"content\": \"This CSV file contains 10 rows and 13 columns. The columns are named 'Year', 'Jan', 'Feb', 'Mar', 'Apr', 'May', 'Jun', 'Jul', 'Aug', 'Sep', 'Oct', 'Nov', 'Dec'. The data types of these columns are int64 for 'Year' and float64 for the rest.\\n\\nIt appears that this CSV file contains monthly inflation rates for different years. The 'Year' column represents the year, and the rest of the columns represent the inflation rates for each month of the year.\", \"role\": \"assistant\", \"stop_reason\": {\"__enum__\": \"StopReason\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"end_of_turn\"}, \"tool_calls\": []}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"UserMessage\", \"data\": {\"content\": \"Plot average yearly inflation as a time series\", \"context\": null, \"role\": \"user\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"CompletionMessage\", \"data\": {\"content\": \"\", \"role\": \"assistant\", \"stop_reason\": {\"__enum__\": \"StopReason\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"end_of_turn\"}, \"tool_calls\": [{\"arguments\": {\"code\": \"import pandas as pd\\nimport matplotlib.pyplot as plt\\n\\n# Load data\\ndf = pd.read_csv(\\\"\")\\n\\n# Calculate average yearly inflation\\ndf['Average'] = df[['Jan', 'Feb', 'Mar', 'Apr', 'May', 'Jun', 'Jul', 'Aug', 'Sep', 'Oct', 'Nov', 'Dec']].mean(axis=1)\\n\\n# Plot time series\\nplt.figure(figsize=(10,6))\\nplt.plot(df['Year'], df['Average'])\\nplt.xlabel('Year')\\nplt.ylabel('Average Yearly Inflation')\\nplt.title('Average Yearly Inflation Over Time')\\nplt.grid(True)\\nplt.show()\"}, \"call_id\": \"\", \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}]}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolResponseMessage\", \"data\": {\"call_id\": \"\", \"content\": \"completed\", \"role\": \"tool\", \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}}]], {\"response_format\": null, \"sampling_params\": {\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"SamplingParams\", \"data\": {\"max_tokens\": 0, \"repetition_penalty\": 1.0, \"strategy\": {\"temperature\": 0.0001, \"top_p\": 0.9, \"type\": \"top_p\"}}}, \"stream\": true, \"tool_config\": {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolConfig\", \"data\": {\"system_message_behavior\": {\"__enum__\": \"SystemMessageBehavior\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"append\"}, \"tool_choice\": {\"__enum__\": \"ToolChoice\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"auto\"}, \"tool_prompt_format\": null}}, \"tool_prompt_format\": null, \"tools\": [{\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"ToolDefinition\", \"data\": {\"description\": \"Execute code\", \"parameters\": {\"code\": {\"default\": null, \"description\": \"The code to execute\", \"param_type\": \"string\", \"required\": true}}, \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}}]}]": { + "chunks": [ + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "start" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "This", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " code will create a line plot of the average", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " yearly inflation over time. The x-axis represents", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " the year, and the y-axis represents the average", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " yearly inflation. The plot will show the trend of average yearly inflation", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " over the years.", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "complete" + }, + "logprobs": null, + "stop_reason": { + "__enum__": "StopReason", + "__module__": "llama_stack.models.llama.datatypes", + "value": "end_of_turn" + } + }, + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 633 + }, + { + "metric": "completion_tokens", + "unit": null, + "value": 56 + }, + { + "metric": "total_tokens", + "unit": null, + "value": 689 + } + ] + } + } + ], + "type": "generator" + }, + "[[\"meta-llama/Llama-3.1-8B-Instruct\", [{\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"SystemMessage\", \"data\": {\"content\": \"You are a helpful assistant\", \"role\": \"system\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"UserMessage\", \"data\": {\"content\": \"Here is a csv, can you describe it?\", \"context\": null, \"role\": \"user\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"CompletionMessage\", \"data\": {\"content\": \"\", \"role\": \"assistant\", \"stop_reason\": {\"__enum__\": \"StopReason\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"end_of_turn\"}, \"tool_calls\": [{\"arguments\": {\"code\": \"import pandas as pd\\n# Load data\\ndf = pd.read_csv(\\\"\")\\n# Rows\\nprint(\\\"Number of rows and columns in the data:\\\", df.shape)\\n# Columns\\nprint(\\\"Columns of the data are:\\\", len(df.columns))\\n# Column names\\nprint(\\\"Columns of the data are:\\\", df.columns)\\n# Column dtypes\\nprint(\\\"Datatype of the columns are:\\\", df.dtypes)\"}, \"call_id\": \"\", \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}]}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolResponseMessage\", \"data\": {\"call_id\": \"\", \"content\": \"completed\\n[stdout]\\nNumber of rows and columns in the data: (10, 13)\\nColumns of the data are: 13\\nColumns of the data are: Index(['Year', 'Jan', 'Feb', 'Mar', 'Apr', 'May', 'Jun', 'Jul', 'Aug', 'Sep',\\n 'Oct', 'Nov', 'Dec'],\\n dtype='object')\\nDatatype of the columns are: Year int64\\nJan float64\\nFeb float64\\nMar float64\\nApr float64\\nMay float64\\nJun float64\\nJul float64\\nAug float64\\nSep float64\\nOct float64\\nNov float64\\nDec float64\\ndtype: object\\n[/stdout]\", \"role\": \"tool\", \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"CompletionMessage\", \"data\": {\"content\": \"This CSV file contains 10 rows and 13 columns. The columns are named 'Year', 'Jan', 'Feb', 'Mar', 'Apr', 'May', 'Jun', 'Jul', 'Aug', 'Sep', 'Oct', 'Nov', 'Dec'. The data types of these columns are int64 for 'Year' and float64 for the rest.\\n\\nIt appears that this CSV file contains monthly inflation rates for different years. The 'Year' column represents the year, and the rest of the columns represent the inflation rates for each month of the year.\", \"role\": \"assistant\", \"stop_reason\": {\"__enum__\": \"StopReason\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"end_of_turn\"}, \"tool_calls\": []}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"UserMessage\", \"data\": {\"content\": \"Plot average yearly inflation as a time series\", \"context\": null, \"role\": \"user\"}}]], {\"response_format\": null, \"sampling_params\": {\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"SamplingParams\", \"data\": {\"max_tokens\": 0, \"repetition_penalty\": 1.0, \"strategy\": {\"temperature\": 0.0001, \"top_p\": 0.9, \"type\": \"top_p\"}}}, \"stream\": true, \"tool_config\": {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolConfig\", \"data\": {\"system_message_behavior\": {\"__enum__\": \"SystemMessageBehavior\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"append\"}, \"tool_choice\": {\"__enum__\": \"ToolChoice\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"auto\"}, \"tool_prompt_format\": null}}, \"tool_prompt_format\": null, \"tools\": [{\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"ToolDefinition\", \"data\": {\"description\": \"Execute code\", \"parameters\": {\"code\": {\"default\": null, \"description\": \"The code to execute\", \"param_type\": \"string\", \"required\": true}}, \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}}]}]": { + "chunks": [ + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "start" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "parse_status": { + "__enum__": "ToolCallParseStatus", + "__module__": "llama_stack.apis.common.content_types", + "value": "started" + }, + "tool_call": "", + "type": "tool_call" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "parse_status": { + "__enum__": "ToolCallParseStatus", + "__module__": "llama_stack.apis.common.content_types", + "value": "in_progress" + }, + "tool_call": "import pandas as pd\nimport matplotlib.pyplot", + "type": "tool_call" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "parse_status": { + "__enum__": "ToolCallParseStatus", + "__module__": "llama_stack.apis.common.content_types", + "value": "in_progress" + }, + "tool_call": " as plt\n\n# Load data\ndf = pd.read_csv(\"/", + "type": "tool_call" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "parse_status": { + "__enum__": "ToolCallParseStatus", + "__module__": "llama_stack.apis.common.content_types", + "value": "in_progress" + }, + "tool_call": "var/folders/cz/vyh7y1d11", + "type": "tool_call" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "parse_status": { + "__enum__": "ToolCallParseStatus", + "__module__": "llama_stack.apis.common.content_types", + "value": "in_progress" + }, + "tool_call": "xg881lsxsshnc5c0000gn/T", + "type": "tool_call" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "parse_status": { + "__enum__": "ToolCallParseStatus", + "__module__": "llama_stack.apis.common.content_types", + "value": "in_progress" + }, + "tool_call": "/tmp_d_cdeif/UuctHlJzinflation.csv\")\n\n", + "type": "tool_call" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "parse_status": { + "__enum__": "ToolCallParseStatus", + "__module__": "llama_stack.apis.common.content_types", + "value": "in_progress" + }, + "tool_call": "# Calculate average yearly inflation\ndf['Average", + "type": "tool_call" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "parse_status": { + "__enum__": "ToolCallParseStatus", + "__module__": "llama_stack.apis.common.content_types", + "value": "in_progress" + }, + "tool_call": "'] = df[['Jan', 'Feb', 'Mar',", + "type": "tool_call" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "parse_status": { + "__enum__": "ToolCallParseStatus", + "__module__": "llama_stack.apis.common.content_types", + "value": "in_progress" + }, + "tool_call": " 'Apr', 'May', 'Jun',", + "type": "tool_call" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "parse_status": { + "__enum__": "ToolCallParseStatus", + "__module__": "llama_stack.apis.common.content_types", + "value": "in_progress" + }, + "tool_call": " 'Jul', 'Aug', 'Sep', 'Oct', '", + "type": "tool_call" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "parse_status": { + "__enum__": "ToolCallParseStatus", + "__module__": "llama_stack.apis.common.content_types", + "value": "in_progress" + }, + "tool_call": "Nov', 'Dec']].mean(axis=1)\n\n# Plot time series", + "type": "tool_call" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "parse_status": { + "__enum__": "ToolCallParseStatus", + "__module__": "llama_stack.apis.common.content_types", + "value": "in_progress" + }, + "tool_call": "\nplt.figure(figsize=(10,6))\nplt.plot(df['Year", + "type": "tool_call" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "parse_status": { + "__enum__": "ToolCallParseStatus", + "__module__": "llama_stack.apis.common.content_types", + "value": "in_progress" + }, + "tool_call": "'], df['Average'])\nplt.xlabel('Year')\nplt.ylabel('Average", + "type": "tool_call" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "parse_status": { + "__enum__": "ToolCallParseStatus", + "__module__": "llama_stack.apis.common.content_types", + "value": "in_progress" + }, + "tool_call": " Yearly Inflation')\nplt.title('Average Yearly Inflation Over", + "type": "tool_call" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "parse_status": { + "__enum__": "ToolCallParseStatus", + "__module__": "llama_stack.apis.common.content_types", + "value": "in_progress" + }, + "tool_call": " Time')\nplt.grid(True)\nplt.show()", + "type": "tool_call" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "parse_status": { + "__enum__": "ToolCallParseStatus", + "__module__": "llama_stack.apis.common.content_types", + "value": "succeeded" + }, + "tool_call": { + "arguments": { + "code": "import pandas as pd\nimport matplotlib.pyplot as plt\n\n# Load data\ndf = pd.read_csv(\"/var/folders/cz/vyh7y1d11xg881lsxsshnc5c0000gn/T/tmp_d_cdeif/UuctHlJzinflation.csv\")\n\n# Calculate average yearly inflation\ndf['Average'] = df[['Jan', 'Feb', 'Mar', 'Apr', 'May', 'Jun', 'Jul', 'Aug', 'Sep', 'Oct', 'Nov', 'Dec']].mean(axis=1)\n\n# Plot time series\nplt.figure(figsize=(10,6))\nplt.plot(df['Year'], df['Average'])\nplt.xlabel('Year')\nplt.ylabel('Average Yearly Inflation')\nplt.title('Average Yearly Inflation Over Time')\nplt.grid(True)\nplt.show()" + }, + "call_id": "f953fd92-9413-4968-9ffa-f85ddea173dc", + "tool_name": { + "__enum__": "BuiltinTool", + "__module__": "llama_stack.models.llama.datatypes", + "value": "code_interpreter" + } + }, + "type": "tool_call" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": { + "__enum__": "StopReason", + "__module__": "llama_stack.models.llama.datatypes", + "value": "end_of_turn" + } + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "complete" + }, + "logprobs": null, + "stop_reason": { + "__enum__": "StopReason", + "__module__": "llama_stack.models.llama.datatypes", + "value": "end_of_turn" + } + }, + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 453 + }, + { + "metric": "completion_tokens", + "unit": null, + "value": 10 + }, + { + "metric": "total_tokens", + "unit": null, + "value": 463 + } + ] + } + } + ], + "type": "generator" + }, "[[\"meta-llama/Llama-3.1-8B-Instruct\", [{\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"SystemMessage\", \"data\": {\"content\": \"You are a helpful assistant\", \"role\": \"system\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"UserMessage\", \"data\": {\"content\": \"Here is a csv, can you describe it?\", \"context\": null, \"role\": \"user\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolResponseMessage\", \"data\": {\"call_id\": \"\", \"content\": [{\"text\": \"# User provided a file accessible to you at \\\"\"\\nYou can use code_interpreter to load and inspect it.\", \"type\": \"text\"}], \"role\": \"tool\", \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"CompletionMessage\", \"data\": {\"content\": \"\", \"role\": \"assistant\", \"stop_reason\": {\"__enum__\": \"StopReason\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"end_of_turn\"}, \"tool_calls\": [{\"arguments\": {\"code\": \"import pandas as pd\\n# Load data\\ndf = pd.read_csv(\\\"\")\\n# Rows\\nprint(\\\"Number of rows and columns in the data:\\\", df.shape)\\n# Columns\\nprint(\\\"Columns of the data are:\\\", len(df.columns))\\n# Column names\\nprint(\\\"Columns of the data are:\\\", df.columns)\\n# Column dtypes\\nprint(\\\"Datatype of the columns are:\\\", df.dtypes)\"}, \"call_id\": \"\", \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}]}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolResponseMessage\", \"data\": {\"call_id\": \"\", \"content\": \"completed\\n[stderr]\\nTraceback (most recent call last):\\n line 5, in \\n from bwrap.core import main\\nModuleNotFoundError: No module named 'bwrap.core'\\n[/stderr]\", \"role\": \"tool\", \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}}]], {\"response_format\": null, \"sampling_params\": {\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"SamplingParams\", \"data\": {\"max_tokens\": 0, \"repetition_penalty\": 1.0, \"strategy\": {\"temperature\": 0.0001, \"top_p\": 0.9, \"type\": \"top_p\"}}}, \"stream\": true, \"tool_config\": {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolConfig\", \"data\": {\"system_message_behavior\": {\"__enum__\": \"SystemMessageBehavior\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"append\"}, \"tool_choice\": {\"__enum__\": \"ToolChoice\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"auto\"}, \"tool_prompt_format\": null}}, \"tool_prompt_format\": null, \"tools\": [{\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"ToolDefinition\", \"data\": {\"description\": \"Execute code\", \"parameters\": {\"code\": {\"default\": null, \"description\": \"The code to execute\", \"param_type\": \"string\", \"required\": true}}, \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}}]}]": { "chunks": [ { @@ -18040,6 +19917,291 @@ ], "type": "generator" }, + "[[\"meta-llama/Llama-3.1-8B-Instruct\", [{\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"SystemMessage\", \"data\": {\"content\": \"You are a helpful assistant\", \"role\": \"system\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"UserMessage\", \"data\": {\"content\": \"Here is a csv, can you describe it?\", \"context\": null, \"role\": \"user\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolResponseMessage\", \"data\": {\"call_id\": \"\", \"content\": [{\"text\": \"# User provided a file accessible to you at \\\"\"\\nYou can use code_interpreter to load and inspect it.\", \"type\": \"text\"}], \"role\": \"tool\", \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"CompletionMessage\", \"data\": {\"content\": \"\", \"role\": \"assistant\", \"stop_reason\": {\"__enum__\": \"StopReason\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"end_of_turn\"}, \"tool_calls\": [{\"arguments\": {\"code\": \"import pandas as pd\\n# Load data\\ndf = pd.read_csv(\\\"\")\\n# Rows\\nprint(\\\"Number of rows and columns in the data:\\\", df.shape)\\n# Columns\\nprint(\\\"Columns of the data are:\\\", len(df.columns))\\n# Column names\\nprint(\\\"Columns of the data are:\\\", df.columns)\\n# Column dtypes\\nprint(\\\"Datatype of the columns are:\\\", df.dtypes)\"}, \"call_id\": \"\", \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}]}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolResponseMessage\", \"data\": {\"call_id\": \"\", \"content\": \"completed\\n[stdout]\\nNumber of rows and columns in the data: (10, 13)\\nColumns of the data are: 13\\nColumns of the data are: Index(['Year', 'Jan', 'Feb', 'Mar', 'Apr', 'May', 'Jun', 'Jul', 'Aug', 'Sep',\\n 'Oct', 'Nov', 'Dec'],\\n dtype='object')\\nDatatype of the columns are: Year int64\\nJan float64\\nFeb float64\\nMar float64\\nApr float64\\nMay float64\\nJun float64\\nJul float64\\nAug float64\\nSep float64\\nOct float64\\nNov float64\\nDec float64\\ndtype: object\\n[/stdout]\", \"role\": \"tool\", \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}}]], {\"response_format\": null, \"sampling_params\": {\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"SamplingParams\", \"data\": {\"max_tokens\": 0, \"repetition_penalty\": 1.0, \"strategy\": {\"temperature\": 0.0001, \"top_p\": 0.9, \"type\": \"top_p\"}}}, \"stream\": true, \"tool_config\": {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolConfig\", \"data\": {\"system_message_behavior\": {\"__enum__\": \"SystemMessageBehavior\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"append\"}, \"tool_choice\": {\"__enum__\": \"ToolChoice\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"auto\"}, \"tool_prompt_format\": null}}, \"tool_prompt_format\": null, \"tools\": [{\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"ToolDefinition\", \"data\": {\"description\": \"Execute code\", \"parameters\": {\"code\": {\"default\": null, \"description\": \"The code to execute\", \"param_type\": \"string\", \"required\": true}}, \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}}]}]": { + "chunks": [ + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "start" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "This", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " CSV file contains 10 rows and 13 columns. The columns", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " are named 'Year', 'Jan', 'Feb', 'Mar',", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " 'Apr', 'May', 'Jun', 'Jul', 'Aug", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "', 'Sep', 'Oct', 'Nov', '", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "Dec'. The data types of these columns are int64 for '", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "Year' and float64 for the rest.\n\nIt appears that this CSV", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " file contains monthly inflation rates for different years. The 'Year' column", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " represents the year, and the rest of the columns represent the inflation rates", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " for each month of the year", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": ".", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "complete" + }, + "logprobs": null, + "stop_reason": { + "__enum__": "StopReason", + "__module__": "llama_stack.models.llama.datatypes", + "value": "end_of_turn" + } + }, + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 326 + }, + { + "metric": "completion_tokens", + "unit": null, + "value": 125 + }, + { + "metric": "total_tokens", + "unit": null, + "value": 451 + } + ] + } + } + ], + "type": "generator" + }, "[[\"meta-llama/Llama-3.1-8B-Instruct\", [{\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"SystemMessage\", \"data\": {\"content\": \"You are a helpful assistant\", \"role\": \"system\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"UserMessage\", \"data\": {\"content\": \"Here is a csv, can you describe it?\", \"context\": null, \"role\": \"user\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolResponseMessage\", \"data\": {\"call_id\": \"\", \"content\": [{\"text\": \"# User provided a file accessible to you at \\\"\"\\nYou can use code_interpreter to load and inspect it.\", \"type\": \"text\"}], \"role\": \"tool\", \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}}]], {\"response_format\": null, \"sampling_params\": {\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"SamplingParams\", \"data\": {\"max_tokens\": 0, \"repetition_penalty\": 1.0, \"strategy\": {\"temperature\": 0.0001, \"top_p\": 0.9, \"type\": \"top_p\"}}}, \"stream\": true, \"tool_config\": {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolConfig\", \"data\": {\"system_message_behavior\": {\"__enum__\": \"SystemMessageBehavior\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"append\"}, \"tool_choice\": {\"__enum__\": \"ToolChoice\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"auto\"}, \"tool_prompt_format\": null}}, \"tool_prompt_format\": null, \"tools\": [{\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"ToolDefinition\", \"data\": {\"description\": \"Execute code\", \"parameters\": {\"code\": {\"default\": null, \"description\": \"The code to execute\", \"param_type\": \"string\", \"required\": true}}, \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}}]}]": { "chunks": [ { @@ -18098,7 +20260,7 @@ "__module__": "llama_stack.apis.common.content_types", "value": "in_progress" }, - "tool_call": "import pandas as pd\n# Load data\ndf = pd.read", + "tool_call": "import pandas as pd\n# Load data\ndf = pd.read_csv(\"/var", "type": "tool_call" }, "event_type": { @@ -18123,7 +20285,7 @@ "__module__": "llama_stack.apis.common.content_types", "value": "in_progress" }, - "tool_call": "_csv(\"/var/folders/cz/vyh7y1d", + "tool_call": "/folders/cz/vyh7y1d11xg881lsx", "type": "tool_call" }, "event_type": { @@ -18148,7 +20310,7 @@ "__module__": "llama_stack.apis.common.content_types", "value": "in_progress" }, - "tool_call": "11xg881lsxsshnc", + "tool_call": "sshnc5c0000gn/T/tmp_d_cdeif/Uuct", "type": "tool_call" }, "event_type": { @@ -18173,7 +20335,7 @@ "__module__": "llama_stack.apis.common.content_types", "value": "in_progress" }, - "tool_call": "5c0000gn/T/tmp4ed7p2bg/U", + "tool_call": "HlJzinflation.csv\")\n# Rows\nprint(\"Number of rows", "type": "tool_call" }, "event_type": { @@ -18198,7 +20360,7 @@ "__module__": "llama_stack.apis.common.content_types", "value": "in_progress" }, - "tool_call": "Z0Z335vinflation.csv\")\n# Rows\nprint(\"", + "tool_call": " and columns in the data:\", df.shape)\n# Columns\nprint(\"Columns", "type": "tool_call" }, "event_type": { @@ -18223,7 +20385,7 @@ "__module__": "llama_stack.apis.common.content_types", "value": "in_progress" }, - "tool_call": "Number of rows and columns in the data:\", df.shape)\n# Columns", + "tool_call": " of the data are:\", len(df.columns))\n# Column names\nprint(\"", "type": "tool_call" }, "event_type": { @@ -18248,7 +20410,7 @@ "__module__": "llama_stack.apis.common.content_types", "value": "in_progress" }, - "tool_call": "\nprint(\"Columns of the data are:\", len(df.columns))\n# Column", + "tool_call": "Columns of the data are:\", df.columns)\n# Column dtypes\nprint", "type": "tool_call" }, "event_type": { @@ -18273,57 +20435,7 @@ "__module__": "llama_stack.apis.common.content_types", "value": "in_progress" }, - "tool_call": " names\nprint(\"Columns of the data are:\", df.columns)\n# Column", - "type": "tool_call" - }, - "event_type": { - "__enum__": "ChatCompletionResponseEventType", - "__module__": "llama_stack.apis.inference.inference", - "value": "progress" - }, - "logprobs": null, - "stop_reason": null - }, - "metrics": null - } - }, - { - "__module__": "llama_stack.apis.inference.inference", - "__pydantic__": "ChatCompletionResponseStreamChunk", - "data": { - "event": { - "delta": { - "parse_status": { - "__enum__": "ToolCallParseStatus", - "__module__": "llama_stack.apis.common.content_types", - "value": "in_progress" - }, - "tool_call": " dtypes\nprint(\"Datatype of the columns are:\", df.dtypes", - "type": "tool_call" - }, - "event_type": { - "__enum__": "ChatCompletionResponseEventType", - "__module__": "llama_stack.apis.inference.inference", - "value": "progress" - }, - "logprobs": null, - "stop_reason": null - }, - "metrics": null - } - }, - { - "__module__": "llama_stack.apis.inference.inference", - "__pydantic__": "ChatCompletionResponseStreamChunk", - "data": { - "event": { - "delta": { - "parse_status": { - "__enum__": "ToolCallParseStatus", - "__module__": "llama_stack.apis.common.content_types", - "value": "in_progress" - }, - "tool_call": ")", + "tool_call": "(\"Datatype of the columns are:\", df.dtypes)", "type": "tool_call" }, "event_type": { @@ -18350,9 +20462,9 @@ }, "tool_call": { "arguments": { - "code": "import pandas as pd\n# Load data\ndf = pd.read_csv(\"/var/folders/cz/vyh7y1d11xg881lsxsshnc5c0000gn/T/tmp4ed7p2bg/UZ0Z335vinflation.csv\")\n# Rows\nprint(\"Number of rows and columns in the data:\", df.shape)\n# Columns\nprint(\"Columns of the data are:\", len(df.columns))\n# Column names\nprint(\"Columns of the data are:\", df.columns)\n# Column dtypes\nprint(\"Datatype of the columns are:\", df.dtypes)" + "code": "import pandas as pd\n# Load data\ndf = pd.read_csv(\"/var/folders/cz/vyh7y1d11xg881lsxsshnc5c0000gn/T/tmp_d_cdeif/UuctHlJzinflation.csv\")\n# Rows\nprint(\"Number of rows and columns in the data:\", df.shape)\n# Columns\nprint(\"Columns of the data are:\", len(df.columns))\n# Column names\nprint(\"Columns of the data are:\", df.columns)\n# Column dtypes\nprint(\"Datatype of the columns are:\", df.dtypes)" }, - "call_id": "98e27ff4-d4d7-4764-9213-f46bb928ec68", + "call_id": "479e0208-711f-4318-b284-745599a9fb9c", "tool_name": { "__enum__": "BuiltinTool", "__module__": "llama_stack.models.llama.datatypes", @@ -18397,8 +20509,1524 @@ "value": "end_of_turn" } }, + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 36 + }, + { + "metric": "completion_tokens", + "unit": null, + "value": 10 + }, + { + "metric": "total_tokens", + "unit": null, + "value": 46 + } + ] + } + } + ], + "type": "generator" + }, + "[[\"meta-llama/Llama-3.1-8B-Instruct\", [{\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"SystemMessage\", \"data\": {\"content\": \"You are a helpful assistant\", \"role\": \"system\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"UserMessage\", \"data\": {\"content\": \"I am attaching some documentation for Torchtune. Help me answer questions I will ask next.\", \"context\": null, \"role\": \"user\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"CompletionMessage\", \"data\": {\"content\": \"\", \"role\": \"assistant\", \"stop_reason\": {\"__enum__\": \"StopReason\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"end_of_turn\"}, \"tool_calls\": [{\"arguments\": {\"query\": \"Torchtune documentation\"}, \"call_id\": \"\", \"tool_name\": \"knowledge_search\"}]}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolResponseMessage\", \"data\": {\"call_id\": \"\", \"content\": [{\"text\": \"knowledge_search tool found 5 chunks:\\nBEGIN of knowledge_search tool results.\\n\", \"type\": \"text\"}, {\"text\": \"Result 1:\\nDocument_id:02bc2\\nContent: conversational data, :func:`~torchtune.datasets.chat_dataset` seems to be a good fit. For any\\ncustom local dataset we always need to specify ``source``, ``data_files``, and ``split`` for any dataset\\nbuilder in torchtune. For :func:`~torchtune.datasets.chat_dataset`, we additionally need to specify\\n``conversation_column`` and ``conversation_style``. Our data follows the ``\\\"sharegpt\\\"`` format, so\\nwe can specify that here. Altogether, our :func:`~torchtune.datasets.chat_dataset` call should\\nlook like so:\\n\\n.. code-block:: python\\n\\n from torchtune.datasets import chat_dataset\\n from torchtune.models.llama3 import llama3_tokenizer\\n\\n tokenizer = llama3_tokenizer(\\\"\")\\n ds = chat_dataset(\\n tokenizer=tokenizer,\\n source=\\\"json\\\",\\n data_files=\\\"data/my_data.json\\\",\\n split=\\\"train\\\",\\n conversation_column=\\\"dialogue\\\",\\n conversation_style=\\\"sharegpt\\\",\\n )\\n\\n.. code-block:: yaml\\n\\n # In config\\n tokenizer:\\n _component_: torchtune.models.llama3.llama3_tokenizer\\n path: dataset:\\n _component_: torchtune.datasets.chat_dataset\\n source: json\\n data_files: data/my_data.json\\n split: train\\n conversation_column: dialogue\\n conversation_style: sharegpt\\n\\n.. note::\\n You can pass in any keyword argument for `load_dataset `_ into all our\\n Dataset classes and they will honor them. This is useful for common parameters\\n such as specifying the data split with :code:`split` or configuration with\\n :code:`name`\\n\\nIf you needed to add a prompt template, you would simply pass it into the tokenizer.\\nSince we're fine-tuning Llama3, the tokenizer will handle all formatting for\\nus and prompt templates are optional. Other models such as Mistral's :class:`~torchtune.models.mistral._tokenizer.MistralTokenizer`,\\nuse a chat template by default (:class:`~torchtune.models.mistral.MistralChatTemplate`) to format\\nall messages according to their `recommendations `_, a parameter-efficient finetuning technique,\\nand show you how you can use torchtune to finetune a Llama2 model with LoRA.\\nIf you already know what LoRA is and want to get straight to running\\nyour own LoRA finetune in torchtune, you can jump to :ref:`LoRA finetuning recipe in torchtune`.\\n\\n.. grid:: 2\\n\\n .. grid-item-card:: :octicon:`mortar-board;1em;` What you will learn\\n\\n * What LoRA is and how it saves memory during finetuning\\n * An overview of LoRA components in torchtune\\n * How to run a LoRA finetune using torchtune\\n * How to experiment with different LoRA configurations\\n\\n .. grid-item-card:: :octicon:`list-unordered;1em;` Prerequisites\\n\\n * Be familiar with :ref:`torchtune`\\n * Make sure to :ref:`install torchtune`\\n * Make sure you have downloaded the :ref:`Llama2-7B model weights`\\n\\nWhat is LoRA?\\n-------------\\n\\n`LoRA `_ is an adapter-based method for\\nparameter-efficient finetuning that adds trainable low-rank decomposition matrices to different layers of a neural network,\\nthen freezes the network's remaining parameters. LoRA is most commonly applied to\\ntransformer models, in which case it is common to add the low-rank matrices\\nto some of the linear projections in each transformer layer's self-attention.\\n\\n.. note::\\n\\n If you're unfamiliar, check out these references for the `definition of rank `_\\n and discussion of `low-rank approximations `_.\\n\\nBy finetuning with LoRA (as opposed to finetuning all model parameters),\\nyou can expect to see memory savings due to a substantial reduction in the\\nnumber of parameters with gradients. When using an optimizer with momentum,\\nlike `AdamW `.\\n.. .. _glossary_fsdp2:\\n\\n\", \"type\": \"text\"}, {\"text\": \"Result 4:\\nDocument_id:e40e6\\nContent: 06% of all params are trainable.\\n\\n.. note::\\n If you are directly using the LoRA recipe (as detailed :ref:`here`), you need only pass the\\n relevant checkpoint path. Loading model weights and setting trainable parameters will be taken care\\n of in the recipe.\\n\\n\\n.. _lora_recipe_label:\\n\\nLoRA finetuning recipe in torchtune\\n-----------------------------------\\n\\nFinally, we can put it all together and finetune a model using torchtune's `LoRA recipe `_.\\nMake sure that you have first downloaded the Llama2 weights and tokenizer by following :ref:`these instructions`.\\nYou can then run the following command to perform a LoRA finetune of Llama2-7B with two GPUs (each having VRAM of at least 16GB):\\n\\n.. code-block:: bash\\n\\n tune run --nnodes 1 --nproc_per_node 2 lora_finetune_distributed --config llama2/7B_lora\\n\\n.. note::\\n Make sure to point to the location of your Llama2 weights and tokenizer. This can be done\\n either by adding :code:`checkpointer.checkpoint_files=[my_model_checkpoint_path] tokenizer_checkpoint=my_tokenizer_checkpoint_path`\\n or by directly modifying the :code:`7B_lora.yaml` file. See our \\\"\\\":ref:`config_tutorial_label`\\\" recipe\\n for more details on how you can easily clone and modify torchtune configs.\\n\\n.. note::\\n You can modify the value of :code:`nproc_per_node` depending on (a) the number of GPUs you have available,\\n and (b) the memory constraints of your hardware.\\n\\nThe preceding command will run a LoRA finetune with torchtune's factory settings, but we may want to experiment a bit.\\nLet's take a closer look at some of the :code:`lora_finetune_distributed` config.\\n\\n.. code-block:: yaml\\n\\n # Model Arguments\\n model:\\n _component_: lora_llama2_7b\\n lora_attn_modules: ['q_proj', 'v_proj']\\n lora_rank: 8\\n lora_alpha: 16\\n ...\\n\\nWe see that the\\n\", \"type\": \"text\"}, {\"text\": \"Result 5:\\nDocument_id:200a9\\nContent: etune\\n:func:`torchtune.models.llama3.llama3_8b` with DoRA, you would use :func:`torchtune.models.llama3.lora_llama3_8b` with ``use_dora=True``:\\n\\n.. code-block:: bash\\n\\n tune run lora_finetune_single_device --config llama3/8B_lora_single_device \\\\\\n model.use_dora=True\\n\\n.. code-block:: yaml\\n\\n model:\\n _component_: torchtune.models.lora_llama3_8b\\n use_dora: True\\n\\nSince DoRA extends LoRA, the parameters for :ref:`customizing LoRA ` are identical. You can also quantize the base model weights like in :ref:`glossary_qlora` by using ``quantize=True`` to reap\\neven more memory savings!\\n\\n.. code-block:: bash\\n\\n tune run lora_finetune_single_device --config llama3/8B_lora_single_device \\\\\\n model.apply_lora_to_mlp=True \\\\\\n model.lora_attn_modules=[\\\"q_proj\\\",\\\"k_proj\\\",\\\"v_proj\\\"] \\\\\\n model.lora_rank=16 \\\\\\n model.lora_alpha=32 \\\\\\n model.use_dora=True \\\\\\n model.quantize_base=True\\n\\n.. code-block:: yaml\\n\\n model:\\n _component_: torchtune.models.lora_llama3_8b\\n apply_lora_to_mlp: True\\n lora_attn_modules: [\\\"q_proj\\\", \\\"k_proj\\\", \\\"v_proj\\\"]\\n lora_rank: 16\\n lora_alpha: 32\\n use_dora: True\\n quantize_base: True\\n\\n\\n.. note::\\n\\n Under the hood, we've enabled DoRA by adding the :class:`~torchtune.modules.peft.DoRALinear` module, which we swap\\n out for :class:`~torchtune.modules.peft.LoRALinear` when ``use_dora=True``.\\n\\n.. _glossary_distrib:\\n\\n\\n.. TODO\\n\\n.. Distributed\\n.. -----------\\n\\n.. .. _glossary_fsdp:\\n\\n.. Fully Sharded Data Parallel (FSDP)\\n.. ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\\n\\n.. All our ``_distributed`` recipes use `FSDP `.\\n.. .. _glossary_fsdp2:\\n\\n\", \"type\": \"text\"}, {\"text\": \"END of knowledge_search tool results.\\n\", \"type\": \"text\"}], \"role\": \"tool\", \"tool_name\": \"knowledge_search\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"CompletionMessage\", \"data\": {\"content\": \"I'm ready to help you answer questions about Torchtune based on the documentation you provided. What's your first question?\", \"role\": \"assistant\", \"stop_reason\": {\"__enum__\": \"StopReason\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"end_of_turn\"}, \"tool_calls\": []}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"UserMessage\", \"data\": {\"content\": \"Tell me how to use LoRA\", \"context\": null, \"role\": \"user\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"CompletionMessage\", \"data\": {\"content\": \"\", \"role\": \"assistant\", \"stop_reason\": {\"__enum__\": \"StopReason\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"end_of_turn\"}, \"tool_calls\": [{\"arguments\": {\"query\": \"How to use LoRA in Torchtune\"}, \"call_id\": \"\", \"tool_name\": \"knowledge_search\"}]}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolResponseMessage\", \"data\": {\"call_id\": \"\", \"content\": [{\"text\": \"knowledge_search tool found 5 chunks:\\nBEGIN of knowledge_search tool results.\\n\", \"type\": \"text\"}, {\"text\": \"Result 1:\\nDocument_id:e40e6\\nContent: .. _lora_finetune_label:\\n\\n============================\\nFine-Tuning Llama2 with LoRA\\n============================\\n\\nThis guide will teach you about `LoRA `_, a parameter-efficient finetuning technique,\\nand show you how you can use torchtune to finetune a Llama2 model with LoRA.\\nIf you already know what LoRA is and want to get straight to running\\nyour own LoRA finetune in torchtune, you can jump to :ref:`LoRA finetuning recipe in torchtune`.\\n\\n.. grid:: 2\\n\\n .. grid-item-card:: :octicon:`mortar-board;1em;` What you will learn\\n\\n * What LoRA is and how it saves memory during finetuning\\n * An overview of LoRA components in torchtune\\n * How to run a LoRA finetune using torchtune\\n * How to experiment with different LoRA configurations\\n\\n .. grid-item-card:: :octicon:`list-unordered;1em;` Prerequisites\\n\\n * Be familiar with :ref:`torchtune`\\n * Make sure to :ref:`install torchtune`\\n * Make sure you have downloaded the :ref:`Llama2-7B model weights`\\n\\nWhat is LoRA?\\n-------------\\n\\n`LoRA `_ is an adapter-based method for\\nparameter-efficient finetuning that adds trainable low-rank decomposition matrices to different layers of a neural network,\\nthen freezes the network's remaining parameters. LoRA is most commonly applied to\\ntransformer models, in which case it is common to add the low-rank matrices\\nto some of the linear projections in each transformer layer's self-attention.\\n\\n.. note::\\n\\n If you're unfamiliar, check out these references for the `definition of rank `_\\n and discussion of `low-rank approximations `_.\\n\\nBy finetuning with LoRA (as opposed to finetuning all model parameters),\\nyou can expect to see memory savings due to a substantial reduction in the\\nnumber of parameters with gradients. When using an optimizer with momentum,\\nlike `AdamW ` alone will not handle the definition of which parameters are trainable.\\n See :ref:`below` for how to do this.\\n\\nLet's inspect each of these models a bit more closely.\\n\\n.. code-block:: bash\\n\\n # Print the first layer's self-attention in the usual Llama2 model\\n >>> print(base_model.layers[0].attn)\\n MultiHeadAttention(\\n (q_proj): Linear(in_features=4096, out_features=4096, bias=False)\\n (k_proj): Linear(in_features=4096, out_features=4096, bias=False)\\n (v_proj): Linear(in_features=4096, out_features=4096, bias=False)\\n (output_proj): Linear(in_features=4096, out_features=4096, bias=False)\\n (pos_embeddings): RotaryPositionalEmbeddings()\\n )\\n\\n # Print the same for Llama2 with LoRA weights\\n >>> print(lora_model.layers[0].attn)\\n MultiHeadAttention(\\n (q_proj): LoRALinear(\\n (dropout): Dropout(p=0.0, inplace=False)\\n \\n\", \"type\": \"text\"}, {\"text\": \"Result 3:\\nDocument_id:e40e6\\nContent: 06% of all params are trainable.\\n\\n.. note::\\n If you are directly using the LoRA recipe (as detailed :ref:`here`), you need only pass the\\n relevant checkpoint path. Loading model weights and setting trainable parameters will be taken care\\n of in the recipe.\\n\\n\\n.. _lora_recipe_label:\\n\\nLoRA finetuning recipe in torchtune\\n-----------------------------------\\n\\nFinally, we can put it all together and finetune a model using torchtune's `LoRA recipe `_.\\nMake sure that you have first downloaded the Llama2 weights and tokenizer by following :ref:`these instructions`.\\nYou can then run the following command to perform a LoRA finetune of Llama2-7B with two GPUs (each having VRAM of at least 16GB):\\n\\n.. code-block:: bash\\n\\n tune run --nnodes 1 --nproc_per_node 2 lora_finetune_distributed --config llama2/7B_lora\\n\\n.. note::\\n Make sure to point to the location of your Llama2 weights and tokenizer. This can be done\\n either by adding :code:`checkpointer.checkpoint_files=[my_model_checkpoint_path] tokenizer_checkpoint=my_tokenizer_checkpoint_path`\\n or by directly modifying the :code:`7B_lora.yaml` file. See our \\\"\\\":ref:`config_tutorial_label`\\\" recipe\\n for more details on how you can easily clone and modify torchtune configs.\\n\\n.. note::\\n You can modify the value of :code:`nproc_per_node` depending on (a) the number of GPUs you have available,\\n and (b) the memory constraints of your hardware.\\n\\nThe preceding command will run a LoRA finetune with torchtune's factory settings, but we may want to experiment a bit.\\nLet's take a closer look at some of the :code:`lora_finetune_distributed` config.\\n\\n.. code-block:: yaml\\n\\n # Model Arguments\\n model:\\n _component_: lora_llama2_7b\\n lora_attn_modules: ['q_proj', 'v_proj']\\n lora_rank: 8\\n lora_alpha: 16\\n ...\\n\\nWe see that the\\n\", \"type\": \"text\"}, {\"text\": \"Result 4:\\nDocument_id:e40e6\\nContent: from our Llama2\\nmodel without any wrappers or custom checkpoint conversion logic.\\n\\n.. code-block:: python\\n\\n # Assuming that base_model already has the pretrained Llama2 weights,\\n # this will directly load them into your LoRA model without any conversion necessary.\\n lora_model.load_state_dict(base_model.state_dict(), strict=False)\\n\\n.. note::\\n Whenever loading weights with :code:`strict=False`, you should verify that any missing or extra keys in\\n the loaded :code:`state_dict` are as expected. torchtune's LoRA recipes do this by default via\\n :func:`validate_missing_and_unexpected_for_lora() `.\\n\\nOnce we've loaded the base model weights, we also want to set only LoRA parameters to trainable.\\n\\n.. _setting_trainable_params:\\n\\n.. code-block:: python\\n\\n from torchtune.modules.peft.peft_utils import get_adapter_params, set_trainable_params\\n\\n # Fetch all params from the model that are associated with LoRA.\\n lora_params = get_adapter_params(lora_model)\\n\\n # Set requires_grad=True on lora_params, and requires_grad=False on all others.\\n set_trainable_params(lora_model, lora_params)\\n\\n # Print the total number of parameters\\n total_params = sum([p.numel() for p in lora_model.parameters()])\\n trainable_params = sum([p.numel() for p in lora_model.parameters() if p.requires_grad])\\n print(\\n f\\\"\\\"\\\"\\n {total_params} total params,\\n {trainable_params}\\\" trainable params,\\n {(100.0 * trainable_params / total_params):.2f}% of all params are trainable.\\n \\\"\\\"\\\"\\n )\\n\\n 6742609920 total params,\\n 4194304 trainable params,\\n 0.06% of all params are trainable.\\n\\n.. note::\\n If you are directly using the LoRA recipe (as detailed :ref:`here`), you need only pass the\\n relevant checkpoint path. Loading model weights and setting trainable parameters will be taken care\\n of in the recipe.\\n\\n\\n.. _lora_recipe_label:\\n\\nLoRA finetuning recipe in torchtune\\n-----------------------------------\\n\\nFinally, we can put it all together and finetune a model using torchtune's `LoRA recipe \", \"tool_name\": \"knowledge_search\"}]}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolResponseMessage\", \"data\": {\"call_id\": \"\", \"content\": [{\"text\": \"knowledge_search tool found 5 chunks:\\nBEGIN of knowledge_search tool results.\\n\", \"type\": \"text\"}, {\"text\": \"Result 1:\\nDocument_id:02bc2\\nContent: conversational data, :func:`~torchtune.datasets.chat_dataset` seems to be a good fit. For any\\ncustom local dataset we always need to specify ``source``, ``data_files``, and ``split`` for any dataset\\nbuilder in torchtune. For :func:`~torchtune.datasets.chat_dataset`, we additionally need to specify\\n``conversation_column`` and ``conversation_style``. Our data follows the ``\\\"sharegpt\\\"`` format, so\\nwe can specify that here. Altogether, our :func:`~torchtune.datasets.chat_dataset` call should\\nlook like so:\\n\\n.. code-block:: python\\n\\n from torchtune.datasets import chat_dataset\\n from torchtune.models.llama3 import llama3_tokenizer\\n\\n tokenizer = llama3_tokenizer(\\\"\")\\n ds = chat_dataset(\\n tokenizer=tokenizer,\\n source=\\\"json\\\",\\n data_files=\\\"data/my_data.json\\\",\\n split=\\\"train\\\",\\n conversation_column=\\\"dialogue\\\",\\n conversation_style=\\\"sharegpt\\\",\\n )\\n\\n.. code-block:: yaml\\n\\n # In config\\n tokenizer:\\n _component_: torchtune.models.llama3.llama3_tokenizer\\n path: dataset:\\n _component_: torchtune.datasets.chat_dataset\\n source: json\\n data_files: data/my_data.json\\n split: train\\n conversation_column: dialogue\\n conversation_style: sharegpt\\n\\n.. note::\\n You can pass in any keyword argument for `load_dataset `_ into all our\\n Dataset classes and they will honor them. This is useful for common parameters\\n such as specifying the data split with :code:`split` or configuration with\\n :code:`name`\\n\\nIf you needed to add a prompt template, you would simply pass it into the tokenizer.\\nSince we're fine-tuning Llama3, the tokenizer will handle all formatting for\\nus and prompt templates are optional. Other models such as Mistral's :class:`~torchtune.models.mistral._tokenizer.MistralTokenizer`,\\nuse a chat template by default (:class:`~torchtune.models.mistral.MistralChatTemplate`) to format\\nall messages according to their `recommendations `_, a parameter-efficient finetuning technique,\\nand show you how you can use torchtune to finetune a Llama2 model with LoRA.\\nIf you already know what LoRA is and want to get straight to running\\nyour own LoRA finetune in torchtune, you can jump to :ref:`LoRA finetuning recipe in torchtune`.\\n\\n.. grid:: 2\\n\\n .. grid-item-card:: :octicon:`mortar-board;1em;` What you will learn\\n\\n * What LoRA is and how it saves memory during finetuning\\n * An overview of LoRA components in torchtune\\n * How to run a LoRA finetune using torchtune\\n * How to experiment with different LoRA configurations\\n\\n .. grid-item-card:: :octicon:`list-unordered;1em;` Prerequisites\\n\\n * Be familiar with :ref:`torchtune`\\n * Make sure to :ref:`install torchtune`\\n * Make sure you have downloaded the :ref:`Llama2-7B model weights`\\n\\nWhat is LoRA?\\n-------------\\n\\n`LoRA `_ is an adapter-based method for\\nparameter-efficient finetuning that adds trainable low-rank decomposition matrices to different layers of a neural network,\\nthen freezes the network's remaining parameters. LoRA is most commonly applied to\\ntransformer models, in which case it is common to add the low-rank matrices\\nto some of the linear projections in each transformer layer's self-attention.\\n\\n.. note::\\n\\n If you're unfamiliar, check out these references for the `definition of rank `_\\n and discussion of `low-rank approximations `_.\\n\\nBy finetuning with LoRA (as opposed to finetuning all model parameters),\\nyou can expect to see memory savings due to a substantial reduction in the\\nnumber of parameters with gradients. When using an optimizer with momentum,\\nlike `AdamW `.\\n.. .. _glossary_fsdp2:\\n\\n\", \"type\": \"text\"}, {\"text\": \"Result 4:\\nDocument_id:e40e6\\nContent: 06% of all params are trainable.\\n\\n.. note::\\n If you are directly using the LoRA recipe (as detailed :ref:`here`), you need only pass the\\n relevant checkpoint path. Loading model weights and setting trainable parameters will be taken care\\n of in the recipe.\\n\\n\\n.. _lora_recipe_label:\\n\\nLoRA finetuning recipe in torchtune\\n-----------------------------------\\n\\nFinally, we can put it all together and finetune a model using torchtune's `LoRA recipe `_.\\nMake sure that you have first downloaded the Llama2 weights and tokenizer by following :ref:`these instructions`.\\nYou can then run the following command to perform a LoRA finetune of Llama2-7B with two GPUs (each having VRAM of at least 16GB):\\n\\n.. code-block:: bash\\n\\n tune run --nnodes 1 --nproc_per_node 2 lora_finetune_distributed --config llama2/7B_lora\\n\\n.. note::\\n Make sure to point to the location of your Llama2 weights and tokenizer. This can be done\\n either by adding :code:`checkpointer.checkpoint_files=[my_model_checkpoint_path] tokenizer_checkpoint=my_tokenizer_checkpoint_path`\\n or by directly modifying the :code:`7B_lora.yaml` file. See our \\\"\\\":ref:`config_tutorial_label`\\\" recipe\\n for more details on how you can easily clone and modify torchtune configs.\\n\\n.. note::\\n You can modify the value of :code:`nproc_per_node` depending on (a) the number of GPUs you have available,\\n and (b) the memory constraints of your hardware.\\n\\nThe preceding command will run a LoRA finetune with torchtune's factory settings, but we may want to experiment a bit.\\nLet's take a closer look at some of the :code:`lora_finetune_distributed` config.\\n\\n.. code-block:: yaml\\n\\n # Model Arguments\\n model:\\n _component_: lora_llama2_7b\\n lora_attn_modules: ['q_proj', 'v_proj']\\n lora_rank: 8\\n lora_alpha: 16\\n ...\\n\\nWe see that the\\n\", \"type\": \"text\"}, {\"text\": \"Result 5:\\nDocument_id:200a9\\nContent: etune\\n:func:`torchtune.models.llama3.llama3_8b` with DoRA, you would use :func:`torchtune.models.llama3.lora_llama3_8b` with ``use_dora=True``:\\n\\n.. code-block:: bash\\n\\n tune run lora_finetune_single_device --config llama3/8B_lora_single_device \\\\\\n model.use_dora=True\\n\\n.. code-block:: yaml\\n\\n model:\\n _component_: torchtune.models.lora_llama3_8b\\n use_dora: True\\n\\nSince DoRA extends LoRA, the parameters for :ref:`customizing LoRA ` are identical. You can also quantize the base model weights like in :ref:`glossary_qlora` by using ``quantize=True`` to reap\\neven more memory savings!\\n\\n.. code-block:: bash\\n\\n tune run lora_finetune_single_device --config llama3/8B_lora_single_device \\\\\\n model.apply_lora_to_mlp=True \\\\\\n model.lora_attn_modules=[\\\"q_proj\\\",\\\"k_proj\\\",\\\"v_proj\\\"] \\\\\\n model.lora_rank=16 \\\\\\n model.lora_alpha=32 \\\\\\n model.use_dora=True \\\\\\n model.quantize_base=True\\n\\n.. code-block:: yaml\\n\\n model:\\n _component_: torchtune.models.lora_llama3_8b\\n apply_lora_to_mlp: True\\n lora_attn_modules: [\\\"q_proj\\\", \\\"k_proj\\\", \\\"v_proj\\\"]\\n lora_rank: 16\\n lora_alpha: 32\\n use_dora: True\\n quantize_base: True\\n\\n\\n.. note::\\n\\n Under the hood, we've enabled DoRA by adding the :class:`~torchtune.modules.peft.DoRALinear` module, which we swap\\n out for :class:`~torchtune.modules.peft.LoRALinear` when ``use_dora=True``.\\n\\n.. _glossary_distrib:\\n\\n\\n.. TODO\\n\\n.. Distributed\\n.. -----------\\n\\n.. .. _glossary_fsdp:\\n\\n.. Fully Sharded Data Parallel (FSDP)\\n.. ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\\n\\n.. All our ``_distributed`` recipes use `FSDP `.\\n.. .. _glossary_fsdp2:\\n\\n\", \"type\": \"text\"}, {\"text\": \"END of knowledge_search tool results.\\n\", \"type\": \"text\"}], \"role\": \"tool\", \"tool_name\": \"knowledge_search\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"CompletionMessage\", \"data\": {\"content\": \"I'm ready to help you answer questions about Torchtune based on the documentation you provided. What's your first question?\", \"role\": \"assistant\", \"stop_reason\": {\"__enum__\": \"StopReason\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"end_of_turn\"}, \"tool_calls\": []}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"UserMessage\", \"data\": {\"content\": \"Tell me how to use LoRA\", \"context\": null, \"role\": \"user\"}}]], {\"response_format\": null, \"sampling_params\": {\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"SamplingParams\", \"data\": {\"max_tokens\": 0, \"repetition_penalty\": 1.0, \"strategy\": {\"temperature\": 0.0001, \"top_p\": 0.9, \"type\": \"top_p\"}}}, \"stream\": true, \"tool_config\": {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolConfig\", \"data\": {\"system_message_behavior\": {\"__enum__\": \"SystemMessageBehavior\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"append\"}, \"tool_choice\": {\"__enum__\": \"ToolChoice\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"auto\"}, \"tool_prompt_format\": null}}, \"tool_prompt_format\": null, \"tools\": [{\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"ToolDefinition\", \"data\": {\"description\": \"Search for information in a database.\", \"parameters\": {\"query\": {\"default\": null, \"description\": \"The query to search for. Can be a natural language sentence or keywords.\", \"param_type\": \"string\", \"required\": true}}, \"tool_name\": \"knowledge_search\"}}]}]": { + "chunks": [ + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "start" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "{\"", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "type\": \"function\", \"name\": \"", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "knowledge_search\", \"parameters\": {\"query\": \"", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "How to use LoRA in Torchtune\"}}", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "parse_status": { + "__enum__": "ToolCallParseStatus", + "__module__": "llama_stack.apis.common.content_types", + "value": "succeeded" + }, + "tool_call": { + "arguments": { + "query": "How to use LoRA in Torchtune" + }, + "call_id": "6ee142d9-1a65-433e-a681-f20066a2e1f7", + "tool_name": "knowledge_search" + }, + "type": "tool_call" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": { + "__enum__": "StopReason", + "__module__": "llama_stack.models.llama.datatypes", + "value": "end_of_turn" + } + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "complete" + }, + "logprobs": null, + "stop_reason": { + "__enum__": "StopReason", + "__module__": "llama_stack.models.llama.datatypes", + "value": "end_of_turn" + } + }, + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 117 + }, + { + "metric": "completion_tokens", + "unit": null, + "value": 40 + }, + { + "metric": "total_tokens", + "unit": null, + "value": 157 + } + ] + } + } + ], + "type": "generator" + }, + "[[\"meta-llama/Llama-3.1-8B-Instruct\", [{\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"SystemMessage\", \"data\": {\"content\": \"You are a helpful assistant\", \"role\": \"system\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"UserMessage\", \"data\": {\"content\": \"I am attaching some documentation for Torchtune. Help me answer questions I will ask next.\", \"context\": null, \"role\": \"user\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"CompletionMessage\", \"data\": {\"content\": \"\", \"role\": \"assistant\", \"stop_reason\": {\"__enum__\": \"StopReason\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"end_of_turn\"}, \"tool_calls\": [{\"arguments\": {\"query\": \"Torchtune documentation\"}, \"call_id\": \"\", \"tool_name\": \"knowledge_search\"}]}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolResponseMessage\", \"data\": {\"call_id\": \"\", \"content\": [{\"text\": \"knowledge_search tool found 5 chunks:\\nBEGIN of knowledge_search tool results.\\n\", \"type\": \"text\"}, {\"text\": \"Result 1:\\nDocument_id:02bc2\\nContent: conversational data, :func:`~torchtune.datasets.chat_dataset` seems to be a good fit. For any\\ncustom local dataset we always need to specify ``source``, ``data_files``, and ``split`` for any dataset\\nbuilder in torchtune. For :func:`~torchtune.datasets.chat_dataset`, we additionally need to specify\\n``conversation_column`` and ``conversation_style``. Our data follows the ``\\\"sharegpt\\\"`` format, so\\nwe can specify that here. Altogether, our :func:`~torchtune.datasets.chat_dataset` call should\\nlook like so:\\n\\n.. code-block:: python\\n\\n from torchtune.datasets import chat_dataset\\n from torchtune.models.llama3 import llama3_tokenizer\\n\\n tokenizer = llama3_tokenizer(\\\"\")\\n ds = chat_dataset(\\n tokenizer=tokenizer,\\n source=\\\"json\\\",\\n data_files=\\\"data/my_data.json\\\",\\n split=\\\"train\\\",\\n conversation_column=\\\"dialogue\\\",\\n conversation_style=\\\"sharegpt\\\",\\n )\\n\\n.. code-block:: yaml\\n\\n # In config\\n tokenizer:\\n _component_: torchtune.models.llama3.llama3_tokenizer\\n path: dataset:\\n _component_: torchtune.datasets.chat_dataset\\n source: json\\n data_files: data/my_data.json\\n split: train\\n conversation_column: dialogue\\n conversation_style: sharegpt\\n\\n.. note::\\n You can pass in any keyword argument for `load_dataset `_ into all our\\n Dataset classes and they will honor them. This is useful for common parameters\\n such as specifying the data split with :code:`split` or configuration with\\n :code:`name`\\n\\nIf you needed to add a prompt template, you would simply pass it into the tokenizer.\\nSince we're fine-tuning Llama3, the tokenizer will handle all formatting for\\nus and prompt templates are optional. Other models such as Mistral's :class:`~torchtune.models.mistral._tokenizer.MistralTokenizer`,\\nuse a chat template by default (:class:`~torchtune.models.mistral.MistralChatTemplate`) to format\\nall messages according to their `recommendations `_, a parameter-efficient finetuning technique,\\nand show you how you can use torchtune to finetune a Llama2 model with LoRA.\\nIf you already know what LoRA is and want to get straight to running\\nyour own LoRA finetune in torchtune, you can jump to :ref:`LoRA finetuning recipe in torchtune`.\\n\\n.. grid:: 2\\n\\n .. grid-item-card:: :octicon:`mortar-board;1em;` What you will learn\\n\\n * What LoRA is and how it saves memory during finetuning\\n * An overview of LoRA components in torchtune\\n * How to run a LoRA finetune using torchtune\\n * How to experiment with different LoRA configurations\\n\\n .. grid-item-card:: :octicon:`list-unordered;1em;` Prerequisites\\n\\n * Be familiar with :ref:`torchtune`\\n * Make sure to :ref:`install torchtune`\\n * Make sure you have downloaded the :ref:`Llama2-7B model weights`\\n\\nWhat is LoRA?\\n-------------\\n\\n`LoRA `_ is an adapter-based method for\\nparameter-efficient finetuning that adds trainable low-rank decomposition matrices to different layers of a neural network,\\nthen freezes the network's remaining parameters. LoRA is most commonly applied to\\ntransformer models, in which case it is common to add the low-rank matrices\\nto some of the linear projections in each transformer layer's self-attention.\\n\\n.. note::\\n\\n If you're unfamiliar, check out these references for the `definition of rank `_\\n and discussion of `low-rank approximations `_.\\n\\nBy finetuning with LoRA (as opposed to finetuning all model parameters),\\nyou can expect to see memory savings due to a substantial reduction in the\\nnumber of parameters with gradients. When using an optimizer with momentum,\\nlike `AdamW `.\\n.. .. _glossary_fsdp2:\\n\\n\", \"type\": \"text\"}, {\"text\": \"Result 4:\\nDocument_id:e40e6\\nContent: 06% of all params are trainable.\\n\\n.. note::\\n If you are directly using the LoRA recipe (as detailed :ref:`here`), you need only pass the\\n relevant checkpoint path. Loading model weights and setting trainable parameters will be taken care\\n of in the recipe.\\n\\n\\n.. _lora_recipe_label:\\n\\nLoRA finetuning recipe in torchtune\\n-----------------------------------\\n\\nFinally, we can put it all together and finetune a model using torchtune's `LoRA recipe `_.\\nMake sure that you have first downloaded the Llama2 weights and tokenizer by following :ref:`these instructions`.\\nYou can then run the following command to perform a LoRA finetune of Llama2-7B with two GPUs (each having VRAM of at least 16GB):\\n\\n.. code-block:: bash\\n\\n tune run --nnodes 1 --nproc_per_node 2 lora_finetune_distributed --config llama2/7B_lora\\n\\n.. note::\\n Make sure to point to the location of your Llama2 weights and tokenizer. This can be done\\n either by adding :code:`checkpointer.checkpoint_files=[my_model_checkpoint_path] tokenizer_checkpoint=my_tokenizer_checkpoint_path`\\n or by directly modifying the :code:`7B_lora.yaml` file. See our \\\"\\\":ref:`config_tutorial_label`\\\" recipe\\n for more details on how you can easily clone and modify torchtune configs.\\n\\n.. note::\\n You can modify the value of :code:`nproc_per_node` depending on (a) the number of GPUs you have available,\\n and (b) the memory constraints of your hardware.\\n\\nThe preceding command will run a LoRA finetune with torchtune's factory settings, but we may want to experiment a bit.\\nLet's take a closer look at some of the :code:`lora_finetune_distributed` config.\\n\\n.. code-block:: yaml\\n\\n # Model Arguments\\n model:\\n _component_: lora_llama2_7b\\n lora_attn_modules: ['q_proj', 'v_proj']\\n lora_rank: 8\\n lora_alpha: 16\\n ...\\n\\nWe see that the\\n\", \"type\": \"text\"}, {\"text\": \"Result 5:\\nDocument_id:200a9\\nContent: etune\\n:func:`torchtune.models.llama3.llama3_8b` with DoRA, you would use :func:`torchtune.models.llama3.lora_llama3_8b` with ``use_dora=True``:\\n\\n.. code-block:: bash\\n\\n tune run lora_finetune_single_device --config llama3/8B_lora_single_device \\\\\\n model.use_dora=True\\n\\n.. code-block:: yaml\\n\\n model:\\n _component_: torchtune.models.lora_llama3_8b\\n use_dora: True\\n\\nSince DoRA extends LoRA, the parameters for :ref:`customizing LoRA ` are identical. You can also quantize the base model weights like in :ref:`glossary_qlora` by using ``quantize=True`` to reap\\neven more memory savings!\\n\\n.. code-block:: bash\\n\\n tune run lora_finetune_single_device --config llama3/8B_lora_single_device \\\\\\n model.apply_lora_to_mlp=True \\\\\\n model.lora_attn_modules=[\\\"q_proj\\\",\\\"k_proj\\\",\\\"v_proj\\\"] \\\\\\n model.lora_rank=16 \\\\\\n model.lora_alpha=32 \\\\\\n model.use_dora=True \\\\\\n model.quantize_base=True\\n\\n.. code-block:: yaml\\n\\n model:\\n _component_: torchtune.models.lora_llama3_8b\\n apply_lora_to_mlp: True\\n lora_attn_modules: [\\\"q_proj\\\", \\\"k_proj\\\", \\\"v_proj\\\"]\\n lora_rank: 16\\n lora_alpha: 32\\n use_dora: True\\n quantize_base: True\\n\\n\\n.. note::\\n\\n Under the hood, we've enabled DoRA by adding the :class:`~torchtune.modules.peft.DoRALinear` module, which we swap\\n out for :class:`~torchtune.modules.peft.LoRALinear` when ``use_dora=True``.\\n\\n.. _glossary_distrib:\\n\\n\\n.. TODO\\n\\n.. Distributed\\n.. -----------\\n\\n.. .. _glossary_fsdp:\\n\\n.. Fully Sharded Data Parallel (FSDP)\\n.. ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\\n\\n.. All our ``_distributed`` recipes use `FSDP `.\\n.. .. _glossary_fsdp2:\\n\\n\", \"type\": \"text\"}, {\"text\": \"END of knowledge_search tool results.\\n\", \"type\": \"text\"}], \"role\": \"tool\", \"tool_name\": \"knowledge_search\"}}]], {\"response_format\": null, \"sampling_params\": {\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"SamplingParams\", \"data\": {\"max_tokens\": 0, \"repetition_penalty\": 1.0, \"strategy\": {\"temperature\": 0.0001, \"top_p\": 0.9, \"type\": \"top_p\"}}}, \"stream\": true, \"tool_config\": {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolConfig\", \"data\": {\"system_message_behavior\": {\"__enum__\": \"SystemMessageBehavior\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"append\"}, \"tool_choice\": {\"__enum__\": \"ToolChoice\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"auto\"}, \"tool_prompt_format\": null}}, \"tool_prompt_format\": null, \"tools\": [{\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"ToolDefinition\", \"data\": {\"description\": \"Search for information in a database.\", \"parameters\": {\"query\": {\"default\": null, \"description\": \"The query to search for. Can be a natural language sentence or keywords.\", \"param_type\": \"string\", \"required\": true}}, \"tool_name\": \"knowledge_search\"}}]}]": { + "chunks": [ + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "start" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "I", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "'m ready to help you answer questions about Torcht", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "une based on the documentation you provided.", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " What's your first question?", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "complete" + }, + "logprobs": null, + "stop_reason": { + "__enum__": "StopReason", + "__module__": "llama_stack.models.llama.datatypes", + "value": "end_of_turn" + } + }, + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 75 + }, + { + "metric": "completion_tokens", + "unit": null, + "value": 35 + }, + { + "metric": "total_tokens", + "unit": null, + "value": 110 + } + ] + } + } + ], + "type": "generator" + }, + "[[\"meta-llama/Llama-3.1-8B-Instruct\", [{\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"SystemMessage\", \"data\": {\"content\": \"You are a helpful assistant\", \"role\": \"system\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"UserMessage\", \"data\": {\"content\": \"I am attaching some documentation for Torchtune. Help me answer questions I will ask next.\", \"context\": null, \"role\": \"user\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"CompletionMessage\", \"data\": {\"content\": \"\", \"role\": \"assistant\", \"stop_reason\": {\"__enum__\": \"StopReason\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"end_of_turn\"}, \"tool_calls\": [{\"arguments\": {\"query\": \"Torchtune documentation\"}, \"call_id\": \"\", \"tool_name\": \"knowledge_search\"}]}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolResponseMessage\", \"data\": {\"call_id\": \"\", \"content\": [{\"text\": \"knowledge_search tool found 5 chunks:\\nBEGIN of knowledge_search tool results.\\n\", \"type\": \"text\"}, {\"text\": \"Result 1:\\nDocument_id:16a6a\\nContent: conversational data, :func:`~torchtune.datasets.chat_dataset` seems to be a good fit. For any\\ncustom local dataset we always need to specify ``source``, ``data_files``, and ``split`` for any dataset\\nbuilder in torchtune. For :func:`~torchtune.datasets.chat_dataset`, we additionally need to specify\\n``conversation_column`` and ``conversation_style``. Our data follows the ``\\\"sharegpt\\\"`` format, so\\nwe can specify that here. Altogether, our :func:`~torchtune.datasets.chat_dataset` call should\\nlook like so:\\n\\n.. code-block:: python\\n\\n from torchtune.datasets import chat_dataset\\n from torchtune.models.llama3 import llama3_tokenizer\\n\\n tokenizer = llama3_tokenizer(\\\"\")\\n ds = chat_dataset(\\n tokenizer=tokenizer,\\n source=\\\"json\\\",\\n data_files=\\\"data/my_data.json\\\",\\n split=\\\"train\\\",\\n conversation_column=\\\"dialogue\\\",\\n conversation_style=\\\"sharegpt\\\",\\n )\\n\\n.. code-block:: yaml\\n\\n # In config\\n tokenizer:\\n _component_: torchtune.models.llama3.llama3_tokenizer\\n path: dataset:\\n _component_: torchtune.datasets.chat_dataset\\n source: json\\n data_files: data/my_data.json\\n split: train\\n conversation_column: dialogue\\n conversation_style: sharegpt\\n\\n.. note::\\n You can pass in any keyword argument for `load_dataset `_ into all our\\n Dataset classes and they will honor them. This is useful for common parameters\\n such as specifying the data split with :code:`split` or configuration with\\n :code:`name`\\n\\nIf you needed to add a prompt template, you would simply pass it into the tokenizer.\\nSince we're fine-tuning Llama3, the tokenizer will handle all formatting for\\nus and prompt templates are optional. Other models such as Mistral's :class:`~torchtune.models.mistral._tokenizer.MistralTokenizer`,\\nuse a chat template by default (:class:`~torchtune.models.mistral.MistralChatTemplate`) to format\\nall messages according to their `recommendations `_, a parameter-efficient finetuning technique,\\nand show you how you can use torchtune to finetune a Llama2 model with LoRA.\\nIf you already know what LoRA is and want to get straight to running\\nyour own LoRA finetune in torchtune, you can jump to :ref:`LoRA finetuning recipe in torchtune`.\\n\\n.. grid:: 2\\n\\n .. grid-item-card:: :octicon:`mortar-board;1em;` What you will learn\\n\\n * What LoRA is and how it saves memory during finetuning\\n * An overview of LoRA components in torchtune\\n * How to run a LoRA finetune using torchtune\\n * How to experiment with different LoRA configurations\\n\\n .. grid-item-card:: :octicon:`list-unordered;1em;` Prerequisites\\n\\n * Be familiar with :ref:`torchtune`\\n * Make sure to :ref:`install torchtune`\\n * Make sure you have downloaded the :ref:`Llama2-7B model weights`\\n\\nWhat is LoRA?\\n-------------\\n\\n`LoRA `_ is an adapter-based method for\\nparameter-efficient finetuning that adds trainable low-rank decomposition matrices to different layers of a neural network,\\nthen freezes the network's remaining parameters. LoRA is most commonly applied to\\ntransformer models, in which case it is common to add the low-rank matrices\\nto some of the linear projections in each transformer layer's self-attention.\\n\\n.. note::\\n\\n If you're unfamiliar, check out these references for the `definition of rank `_\\n and discussion of `low-rank approximations `_.\\n\\nBy finetuning with LoRA (as opposed to finetuning all model parameters),\\nyou can expect to see memory savings due to a substantial reduction in the\\nnumber of parameters with gradients. When using an optimizer with momentum,\\nlike `AdamW `.\\n.. .. _glossary_fsdp2:\\n\\n\", \"type\": \"text\"}, {\"text\": \"Result 4:\\nDocument_id:cc255\\nContent: 06% of all params are trainable.\\n\\n.. note::\\n If you are directly using the LoRA recipe (as detailed :ref:`here`), you need only pass the\\n relevant checkpoint path. Loading model weights and setting trainable parameters will be taken care\\n of in the recipe.\\n\\n\\n.. _lora_recipe_label:\\n\\nLoRA finetuning recipe in torchtune\\n-----------------------------------\\n\\nFinally, we can put it all together and finetune a model using torchtune's `LoRA recipe `_.\\nMake sure that you have first downloaded the Llama2 weights and tokenizer by following :ref:`these instructions`.\\nYou can then run the following command to perform a LoRA finetune of Llama2-7B with two GPUs (each having VRAM of at least 16GB):\\n\\n.. code-block:: bash\\n\\n tune run --nnodes 1 --nproc_per_node 2 lora_finetune_distributed --config llama2/7B_lora\\n\\n.. note::\\n Make sure to point to the location of your Llama2 weights and tokenizer. This can be done\\n either by adding :code:`checkpointer.checkpoint_files=[my_model_checkpoint_path] tokenizer_checkpoint=my_tokenizer_checkpoint_path`\\n or by directly modifying the :code:`7B_lora.yaml` file. See our \\\"\\\":ref:`config_tutorial_label`\\\" recipe\\n for more details on how you can easily clone and modify torchtune configs.\\n\\n.. note::\\n You can modify the value of :code:`nproc_per_node` depending on (a) the number of GPUs you have available,\\n and (b) the memory constraints of your hardware.\\n\\nThe preceding command will run a LoRA finetune with torchtune's factory settings, but we may want to experiment a bit.\\nLet's take a closer look at some of the :code:`lora_finetune_distributed` config.\\n\\n.. code-block:: yaml\\n\\n # Model Arguments\\n model:\\n _component_: lora_llama2_7b\\n lora_attn_modules: ['q_proj', 'v_proj']\\n lora_rank: 8\\n lora_alpha: 16\\n ...\\n\\nWe see that the\\n\", \"type\": \"text\"}, {\"text\": \"Result 5:\\nDocument_id:7a06a\\nContent: etune\\n:func:`torchtune.models.llama3.llama3_8b` with DoRA, you would use :func:`torchtune.models.llama3.lora_llama3_8b` with ``use_dora=True``:\\n\\n.. code-block:: bash\\n\\n tune run lora_finetune_single_device --config llama3/8B_lora_single_device \\\\\\n model.use_dora=True\\n\\n.. code-block:: yaml\\n\\n model:\\n _component_: torchtune.models.lora_llama3_8b\\n use_dora: True\\n\\nSince DoRA extends LoRA, the parameters for :ref:`customizing LoRA ` are identical. You can also quantize the base model weights like in :ref:`glossary_qlora` by using ``quantize=True`` to reap\\neven more memory savings!\\n\\n.. code-block:: bash\\n\\n tune run lora_finetune_single_device --config llama3/8B_lora_single_device \\\\\\n model.apply_lora_to_mlp=True \\\\\\n model.lora_attn_modules=[\\\"q_proj\\\",\\\"k_proj\\\",\\\"v_proj\\\"] \\\\\\n model.lora_rank=16 \\\\\\n model.lora_alpha=32 \\\\\\n model.use_dora=True \\\\\\n model.quantize_base=True\\n\\n.. code-block:: yaml\\n\\n model:\\n _component_: torchtune.models.lora_llama3_8b\\n apply_lora_to_mlp: True\\n lora_attn_modules: [\\\"q_proj\\\", \\\"k_proj\\\", \\\"v_proj\\\"]\\n lora_rank: 16\\n lora_alpha: 32\\n use_dora: True\\n quantize_base: True\\n\\n\\n.. note::\\n\\n Under the hood, we've enabled DoRA by adding the :class:`~torchtune.modules.peft.DoRALinear` module, which we swap\\n out for :class:`~torchtune.modules.peft.LoRALinear` when ``use_dora=True``.\\n\\n.. _glossary_distrib:\\n\\n\\n.. TODO\\n\\n.. Distributed\\n.. -----------\\n\\n.. .. _glossary_fsdp:\\n\\n.. Fully Sharded Data Parallel (FSDP)\\n.. ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\\n\\n.. All our ``_distributed`` recipes use `FSDP `.\\n.. .. _glossary_fsdp2:\\n\\n\", \"type\": \"text\"}, {\"text\": \"END of knowledge_search tool results.\\n\", \"type\": \"text\"}], \"role\": \"tool\", \"tool_name\": \"knowledge_search\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"CompletionMessage\", \"data\": {\"content\": \"I'm ready to help you answer questions about Torchtune based on the documentation you provided. What's your first question?\", \"role\": \"assistant\", \"stop_reason\": {\"__enum__\": \"StopReason\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"end_of_turn\"}, \"tool_calls\": []}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"UserMessage\", \"data\": {\"content\": \"Tell me how to use LoRA\", \"context\": null, \"role\": \"user\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"CompletionMessage\", \"data\": {\"content\": \"\", \"role\": \"assistant\", \"stop_reason\": {\"__enum__\": \"StopReason\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"end_of_turn\"}, \"tool_calls\": [{\"arguments\": {\"query\": \"How to use LoRA in Torchtune\"}, \"call_id\": \"\", \"tool_name\": \"knowledge_search\"}]}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolResponseMessage\", \"data\": {\"call_id\": \"\", \"content\": [{\"text\": \"knowledge_search tool found 5 chunks:\\nBEGIN of knowledge_search tool results.\\n\", \"type\": \"text\"}, {\"text\": \"Result 1:\\nDocument_id:cc255\\nContent: .. _lora_finetune_label:\\n\\n============================\\nFine-Tuning Llama2 with LoRA\\n============================\\n\\nThis guide will teach you about `LoRA `_, a parameter-efficient finetuning technique,\\nand show you how you can use torchtune to finetune a Llama2 model with LoRA.\\nIf you already know what LoRA is and want to get straight to running\\nyour own LoRA finetune in torchtune, you can jump to :ref:`LoRA finetuning recipe in torchtune`.\\n\\n.. grid:: 2\\n\\n .. grid-item-card:: :octicon:`mortar-board;1em;` What you will learn\\n\\n * What LoRA is and how it saves memory during finetuning\\n * An overview of LoRA components in torchtune\\n * How to run a LoRA finetune using torchtune\\n * How to experiment with different LoRA configurations\\n\\n .. grid-item-card:: :octicon:`list-unordered;1em;` Prerequisites\\n\\n * Be familiar with :ref:`torchtune`\\n * Make sure to :ref:`install torchtune`\\n * Make sure you have downloaded the :ref:`Llama2-7B model weights`\\n\\nWhat is LoRA?\\n-------------\\n\\n`LoRA `_ is an adapter-based method for\\nparameter-efficient finetuning that adds trainable low-rank decomposition matrices to different layers of a neural network,\\nthen freezes the network's remaining parameters. LoRA is most commonly applied to\\ntransformer models, in which case it is common to add the low-rank matrices\\nto some of the linear projections in each transformer layer's self-attention.\\n\\n.. note::\\n\\n If you're unfamiliar, check out these references for the `definition of rank `_\\n and discussion of `low-rank approximations `_.\\n\\nBy finetuning with LoRA (as opposed to finetuning all model parameters),\\nyou can expect to see memory savings due to a substantial reduction in the\\nnumber of parameters with gradients. When using an optimizer with momentum,\\nlike `AdamW ` alone will not handle the definition of which parameters are trainable.\\n See :ref:`below` for how to do this.\\n\\nLet's inspect each of these models a bit more closely.\\n\\n.. code-block:: bash\\n\\n # Print the first layer's self-attention in the usual Llama2 model\\n >>> print(base_model.layers[0].attn)\\n MultiHeadAttention(\\n (q_proj): Linear(in_features=4096, out_features=4096, bias=False)\\n (k_proj): Linear(in_features=4096, out_features=4096, bias=False)\\n (v_proj): Linear(in_features=4096, out_features=4096, bias=False)\\n (output_proj): Linear(in_features=4096, out_features=4096, bias=False)\\n (pos_embeddings): RotaryPositionalEmbeddings()\\n )\\n\\n # Print the same for Llama2 with LoRA weights\\n >>> print(lora_model.layers[0].attn)\\n MultiHeadAttention(\\n (q_proj): LoRALinear(\\n (dropout): Dropout(p=0.0, inplace=False)\\n \\n\", \"type\": \"text\"}, {\"text\": \"Result 3:\\nDocument_id:cc255\\nContent: 06% of all params are trainable.\\n\\n.. note::\\n If you are directly using the LoRA recipe (as detailed :ref:`here`), you need only pass the\\n relevant checkpoint path. Loading model weights and setting trainable parameters will be taken care\\n of in the recipe.\\n\\n\\n.. _lora_recipe_label:\\n\\nLoRA finetuning recipe in torchtune\\n-----------------------------------\\n\\nFinally, we can put it all together and finetune a model using torchtune's `LoRA recipe `_.\\nMake sure that you have first downloaded the Llama2 weights and tokenizer by following :ref:`these instructions`.\\nYou can then run the following command to perform a LoRA finetune of Llama2-7B with two GPUs (each having VRAM of at least 16GB):\\n\\n.. code-block:: bash\\n\\n tune run --nnodes 1 --nproc_per_node 2 lora_finetune_distributed --config llama2/7B_lora\\n\\n.. note::\\n Make sure to point to the location of your Llama2 weights and tokenizer. This can be done\\n either by adding :code:`checkpointer.checkpoint_files=[my_model_checkpoint_path] tokenizer_checkpoint=my_tokenizer_checkpoint_path`\\n or by directly modifying the :code:`7B_lora.yaml` file. See our \\\"\\\":ref:`config_tutorial_label`\\\" recipe\\n for more details on how you can easily clone and modify torchtune configs.\\n\\n.. note::\\n You can modify the value of :code:`nproc_per_node` depending on (a) the number of GPUs you have available,\\n and (b) the memory constraints of your hardware.\\n\\nThe preceding command will run a LoRA finetune with torchtune's factory settings, but we may want to experiment a bit.\\nLet's take a closer look at some of the :code:`lora_finetune_distributed` config.\\n\\n.. code-block:: yaml\\n\\n # Model Arguments\\n model:\\n _component_: lora_llama2_7b\\n lora_attn_modules: ['q_proj', 'v_proj']\\n lora_rank: 8\\n lora_alpha: 16\\n ...\\n\\nWe see that the\\n\", \"type\": \"text\"}, {\"text\": \"Result 4:\\nDocument_id:cc255\\nContent: from our Llama2\\nmodel without any wrappers or custom checkpoint conversion logic.\\n\\n.. code-block:: python\\n\\n # Assuming that base_model already has the pretrained Llama2 weights,\\n # this will directly load them into your LoRA model without any conversion necessary.\\n lora_model.load_state_dict(base_model.state_dict(), strict=False)\\n\\n.. note::\\n Whenever loading weights with :code:`strict=False`, you should verify that any missing or extra keys in\\n the loaded :code:`state_dict` are as expected. torchtune's LoRA recipes do this by default via\\n :func:`validate_missing_and_unexpected_for_lora() `.\\n\\nOnce we've loaded the base model weights, we also want to set only LoRA parameters to trainable.\\n\\n.. _setting_trainable_params:\\n\\n.. code-block:: python\\n\\n from torchtune.modules.peft.peft_utils import get_adapter_params, set_trainable_params\\n\\n # Fetch all params from the model that are associated with LoRA.\\n lora_params = get_adapter_params(lora_model)\\n\\n # Set requires_grad=True on lora_params, and requires_grad=False on all others.\\n set_trainable_params(lora_model, lora_params)\\n\\n # Print the total number of parameters\\n total_params = sum([p.numel() for p in lora_model.parameters()])\\n trainable_params = sum([p.numel() for p in lora_model.parameters() if p.requires_grad])\\n print(\\n f\\\"\\\"\\\"\\n {total_params} total params,\\n {trainable_params}\\\" trainable params,\\n {(100.0 * trainable_params / total_params):.2f}% of all params are trainable.\\n \\\"\\\"\\\"\\n )\\n\\n 6742609920 total params,\\n 4194304 trainable params,\\n 0.06% of all params are trainable.\\n\\n.. note::\\n If you are directly using the LoRA recipe (as detailed :ref:`here`), you need only pass the\\n relevant checkpoint path. Loading model weights and setting trainable parameters will be taken care\\n of in the recipe.\\n\\n\\n.. _lora_recipe_label:\\n\\nLoRA finetuning recipe in torchtune\\n-----------------------------------\\n\\nFinally, we can put it all together and finetune a model using torchtune's `LoRA recipe \", \"tool_name\": \"knowledge_search\"}]}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolResponseMessage\", \"data\": {\"call_id\": \"\", \"content\": [{\"text\": \"knowledge_search tool found 5 chunks:\\nBEGIN of knowledge_search tool results.\\n\", \"type\": \"text\"}, {\"text\": \"Result 1:\\nDocument_id:16a6a\\nContent: conversational data, :func:`~torchtune.datasets.chat_dataset` seems to be a good fit. For any\\ncustom local dataset we always need to specify ``source``, ``data_files``, and ``split`` for any dataset\\nbuilder in torchtune. For :func:`~torchtune.datasets.chat_dataset`, we additionally need to specify\\n``conversation_column`` and ``conversation_style``. Our data follows the ``\\\"sharegpt\\\"`` format, so\\nwe can specify that here. Altogether, our :func:`~torchtune.datasets.chat_dataset` call should\\nlook like so:\\n\\n.. code-block:: python\\n\\n from torchtune.datasets import chat_dataset\\n from torchtune.models.llama3 import llama3_tokenizer\\n\\n tokenizer = llama3_tokenizer(\\\"\")\\n ds = chat_dataset(\\n tokenizer=tokenizer,\\n source=\\\"json\\\",\\n data_files=\\\"data/my_data.json\\\",\\n split=\\\"train\\\",\\n conversation_column=\\\"dialogue\\\",\\n conversation_style=\\\"sharegpt\\\",\\n )\\n\\n.. code-block:: yaml\\n\\n # In config\\n tokenizer:\\n _component_: torchtune.models.llama3.llama3_tokenizer\\n path: dataset:\\n _component_: torchtune.datasets.chat_dataset\\n source: json\\n data_files: data/my_data.json\\n split: train\\n conversation_column: dialogue\\n conversation_style: sharegpt\\n\\n.. note::\\n You can pass in any keyword argument for `load_dataset `_ into all our\\n Dataset classes and they will honor them. This is useful for common parameters\\n such as specifying the data split with :code:`split` or configuration with\\n :code:`name`\\n\\nIf you needed to add a prompt template, you would simply pass it into the tokenizer.\\nSince we're fine-tuning Llama3, the tokenizer will handle all formatting for\\nus and prompt templates are optional. Other models such as Mistral's :class:`~torchtune.models.mistral._tokenizer.MistralTokenizer`,\\nuse a chat template by default (:class:`~torchtune.models.mistral.MistralChatTemplate`) to format\\nall messages according to their `recommendations `_, a parameter-efficient finetuning technique,\\nand show you how you can use torchtune to finetune a Llama2 model with LoRA.\\nIf you already know what LoRA is and want to get straight to running\\nyour own LoRA finetune in torchtune, you can jump to :ref:`LoRA finetuning recipe in torchtune`.\\n\\n.. grid:: 2\\n\\n .. grid-item-card:: :octicon:`mortar-board;1em;` What you will learn\\n\\n * What LoRA is and how it saves memory during finetuning\\n * An overview of LoRA components in torchtune\\n * How to run a LoRA finetune using torchtune\\n * How to experiment with different LoRA configurations\\n\\n .. grid-item-card:: :octicon:`list-unordered;1em;` Prerequisites\\n\\n * Be familiar with :ref:`torchtune`\\n * Make sure to :ref:`install torchtune`\\n * Make sure you have downloaded the :ref:`Llama2-7B model weights`\\n\\nWhat is LoRA?\\n-------------\\n\\n`LoRA `_ is an adapter-based method for\\nparameter-efficient finetuning that adds trainable low-rank decomposition matrices to different layers of a neural network,\\nthen freezes the network's remaining parameters. LoRA is most commonly applied to\\ntransformer models, in which case it is common to add the low-rank matrices\\nto some of the linear projections in each transformer layer's self-attention.\\n\\n.. note::\\n\\n If you're unfamiliar, check out these references for the `definition of rank `_\\n and discussion of `low-rank approximations `_.\\n\\nBy finetuning with LoRA (as opposed to finetuning all model parameters),\\nyou can expect to see memory savings due to a substantial reduction in the\\nnumber of parameters with gradients. When using an optimizer with momentum,\\nlike `AdamW `.\\n.. .. _glossary_fsdp2:\\n\\n\", \"type\": \"text\"}, {\"text\": \"Result 4:\\nDocument_id:cc255\\nContent: 06% of all params are trainable.\\n\\n.. note::\\n If you are directly using the LoRA recipe (as detailed :ref:`here`), you need only pass the\\n relevant checkpoint path. Loading model weights and setting trainable parameters will be taken care\\n of in the recipe.\\n\\n\\n.. _lora_recipe_label:\\n\\nLoRA finetuning recipe in torchtune\\n-----------------------------------\\n\\nFinally, we can put it all together and finetune a model using torchtune's `LoRA recipe `_.\\nMake sure that you have first downloaded the Llama2 weights and tokenizer by following :ref:`these instructions`.\\nYou can then run the following command to perform a LoRA finetune of Llama2-7B with two GPUs (each having VRAM of at least 16GB):\\n\\n.. code-block:: bash\\n\\n tune run --nnodes 1 --nproc_per_node 2 lora_finetune_distributed --config llama2/7B_lora\\n\\n.. note::\\n Make sure to point to the location of your Llama2 weights and tokenizer. This can be done\\n either by adding :code:`checkpointer.checkpoint_files=[my_model_checkpoint_path] tokenizer_checkpoint=my_tokenizer_checkpoint_path`\\n or by directly modifying the :code:`7B_lora.yaml` file. See our \\\"\\\":ref:`config_tutorial_label`\\\" recipe\\n for more details on how you can easily clone and modify torchtune configs.\\n\\n.. note::\\n You can modify the value of :code:`nproc_per_node` depending on (a) the number of GPUs you have available,\\n and (b) the memory constraints of your hardware.\\n\\nThe preceding command will run a LoRA finetune with torchtune's factory settings, but we may want to experiment a bit.\\nLet's take a closer look at some of the :code:`lora_finetune_distributed` config.\\n\\n.. code-block:: yaml\\n\\n # Model Arguments\\n model:\\n _component_: lora_llama2_7b\\n lora_attn_modules: ['q_proj', 'v_proj']\\n lora_rank: 8\\n lora_alpha: 16\\n ...\\n\\nWe see that the\\n\", \"type\": \"text\"}, {\"text\": \"Result 5:\\nDocument_id:7a06a\\nContent: etune\\n:func:`torchtune.models.llama3.llama3_8b` with DoRA, you would use :func:`torchtune.models.llama3.lora_llama3_8b` with ``use_dora=True``:\\n\\n.. code-block:: bash\\n\\n tune run lora_finetune_single_device --config llama3/8B_lora_single_device \\\\\\n model.use_dora=True\\n\\n.. code-block:: yaml\\n\\n model:\\n _component_: torchtune.models.lora_llama3_8b\\n use_dora: True\\n\\nSince DoRA extends LoRA, the parameters for :ref:`customizing LoRA ` are identical. You can also quantize the base model weights like in :ref:`glossary_qlora` by using ``quantize=True`` to reap\\neven more memory savings!\\n\\n.. code-block:: bash\\n\\n tune run lora_finetune_single_device --config llama3/8B_lora_single_device \\\\\\n model.apply_lora_to_mlp=True \\\\\\n model.lora_attn_modules=[\\\"q_proj\\\",\\\"k_proj\\\",\\\"v_proj\\\"] \\\\\\n model.lora_rank=16 \\\\\\n model.lora_alpha=32 \\\\\\n model.use_dora=True \\\\\\n model.quantize_base=True\\n\\n.. code-block:: yaml\\n\\n model:\\n _component_: torchtune.models.lora_llama3_8b\\n apply_lora_to_mlp: True\\n lora_attn_modules: [\\\"q_proj\\\", \\\"k_proj\\\", \\\"v_proj\\\"]\\n lora_rank: 16\\n lora_alpha: 32\\n use_dora: True\\n quantize_base: True\\n\\n\\n.. note::\\n\\n Under the hood, we've enabled DoRA by adding the :class:`~torchtune.modules.peft.DoRALinear` module, which we swap\\n out for :class:`~torchtune.modules.peft.LoRALinear` when ``use_dora=True``.\\n\\n.. _glossary_distrib:\\n\\n\\n.. TODO\\n\\n.. Distributed\\n.. -----------\\n\\n.. .. _glossary_fsdp:\\n\\n.. Fully Sharded Data Parallel (FSDP)\\n.. ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\\n\\n.. All our ``_distributed`` recipes use `FSDP `.\\n.. .. _glossary_fsdp2:\\n\\n\", \"type\": \"text\"}, {\"text\": \"END of knowledge_search tool results.\\n\", \"type\": \"text\"}], \"role\": \"tool\", \"tool_name\": \"knowledge_search\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"CompletionMessage\", \"data\": {\"content\": \"I'm ready to help you answer questions about Torchtune based on the documentation you provided. What's your first question?\", \"role\": \"assistant\", \"stop_reason\": {\"__enum__\": \"StopReason\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"end_of_turn\"}, \"tool_calls\": []}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"UserMessage\", \"data\": {\"content\": \"Tell me how to use LoRA\", \"context\": null, \"role\": \"user\"}}]], {\"response_format\": null, \"sampling_params\": {\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"SamplingParams\", \"data\": {\"max_tokens\": 0, \"repetition_penalty\": 1.0, \"strategy\": {\"temperature\": 0.0001, \"top_p\": 0.9, \"type\": \"top_p\"}}}, \"stream\": true, \"tool_config\": {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolConfig\", \"data\": {\"system_message_behavior\": {\"__enum__\": \"SystemMessageBehavior\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"append\"}, \"tool_choice\": {\"__enum__\": \"ToolChoice\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"auto\"}, \"tool_prompt_format\": null}}, \"tool_prompt_format\": null, \"tools\": [{\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"ToolDefinition\", \"data\": {\"description\": \"Search for information in a database.\", \"parameters\": {\"query\": {\"default\": null, \"description\": \"The query to search for. Can be a natural language sentence or keywords.\", \"param_type\": \"string\", \"required\": true}}, \"tool_name\": \"knowledge_search\"}}]}]": { + "chunks": [ + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "start" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "{\"", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "type\": \"function\", \"name\": \"knowledge_search\", \"parameters", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "\": {\"query\": \"How to use LoRA in Torchtune", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "\"}}", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "parse_status": { + "__enum__": "ToolCallParseStatus", + "__module__": "llama_stack.apis.common.content_types", + "value": "succeeded" + }, + "tool_call": { + "arguments": { + "query": "How to use LoRA in Torchtune" + }, + "call_id": "a7b02498-0a50-40c2-abf2-563d4d26d01f", + "tool_name": "knowledge_search" + }, + "type": "tool_call" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": { + "__enum__": "StopReason", + "__module__": "llama_stack.models.llama.datatypes", + "value": "end_of_turn" + } + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "complete" + }, + "logprobs": null, + "stop_reason": { + "__enum__": "StopReason", + "__module__": "llama_stack.models.llama.datatypes", + "value": "end_of_turn" + } + }, + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 117 + }, + { + "metric": "completion_tokens", + "unit": null, + "value": 40 + }, + { + "metric": "total_tokens", + "unit": null, + "value": 157 + } + ] + } + } + ], + "type": "generator" + }, + "[[\"meta-llama/Llama-3.1-8B-Instruct\", [{\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"SystemMessage\", \"data\": {\"content\": \"You are a helpful assistant\", \"role\": \"system\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"UserMessage\", \"data\": {\"content\": \"I am attaching some documentation for Torchtune. Help me answer questions I will ask next.\", \"context\": null, \"role\": \"user\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"CompletionMessage\", \"data\": {\"content\": \"\", \"role\": \"assistant\", \"stop_reason\": {\"__enum__\": \"StopReason\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"end_of_turn\"}, \"tool_calls\": [{\"arguments\": {\"query\": \"Torchtune documentation\"}, \"call_id\": \"\", \"tool_name\": \"knowledge_search\"}]}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolResponseMessage\", \"data\": {\"call_id\": \"\", \"content\": [{\"text\": \"knowledge_search tool found 5 chunks:\\nBEGIN of knowledge_search tool results.\\n\", \"type\": \"text\"}, {\"text\": \"Result 1:\\nDocument_id:16a6a\\nContent: conversational data, :func:`~torchtune.datasets.chat_dataset` seems to be a good fit. For any\\ncustom local dataset we always need to specify ``source``, ``data_files``, and ``split`` for any dataset\\nbuilder in torchtune. For :func:`~torchtune.datasets.chat_dataset`, we additionally need to specify\\n``conversation_column`` and ``conversation_style``. Our data follows the ``\\\"sharegpt\\\"`` format, so\\nwe can specify that here. Altogether, our :func:`~torchtune.datasets.chat_dataset` call should\\nlook like so:\\n\\n.. code-block:: python\\n\\n from torchtune.datasets import chat_dataset\\n from torchtune.models.llama3 import llama3_tokenizer\\n\\n tokenizer = llama3_tokenizer(\\\"\")\\n ds = chat_dataset(\\n tokenizer=tokenizer,\\n source=\\\"json\\\",\\n data_files=\\\"data/my_data.json\\\",\\n split=\\\"train\\\",\\n conversation_column=\\\"dialogue\\\",\\n conversation_style=\\\"sharegpt\\\",\\n )\\n\\n.. code-block:: yaml\\n\\n # In config\\n tokenizer:\\n _component_: torchtune.models.llama3.llama3_tokenizer\\n path: dataset:\\n _component_: torchtune.datasets.chat_dataset\\n source: json\\n data_files: data/my_data.json\\n split: train\\n conversation_column: dialogue\\n conversation_style: sharegpt\\n\\n.. note::\\n You can pass in any keyword argument for `load_dataset `_ into all our\\n Dataset classes and they will honor them. This is useful for common parameters\\n such as specifying the data split with :code:`split` or configuration with\\n :code:`name`\\n\\nIf you needed to add a prompt template, you would simply pass it into the tokenizer.\\nSince we're fine-tuning Llama3, the tokenizer will handle all formatting for\\nus and prompt templates are optional. Other models such as Mistral's :class:`~torchtune.models.mistral._tokenizer.MistralTokenizer`,\\nuse a chat template by default (:class:`~torchtune.models.mistral.MistralChatTemplate`) to format\\nall messages according to their `recommendations `_, a parameter-efficient finetuning technique,\\nand show you how you can use torchtune to finetune a Llama2 model with LoRA.\\nIf you already know what LoRA is and want to get straight to running\\nyour own LoRA finetune in torchtune, you can jump to :ref:`LoRA finetuning recipe in torchtune`.\\n\\n.. grid:: 2\\n\\n .. grid-item-card:: :octicon:`mortar-board;1em;` What you will learn\\n\\n * What LoRA is and how it saves memory during finetuning\\n * An overview of LoRA components in torchtune\\n * How to run a LoRA finetune using torchtune\\n * How to experiment with different LoRA configurations\\n\\n .. grid-item-card:: :octicon:`list-unordered;1em;` Prerequisites\\n\\n * Be familiar with :ref:`torchtune`\\n * Make sure to :ref:`install torchtune`\\n * Make sure you have downloaded the :ref:`Llama2-7B model weights`\\n\\nWhat is LoRA?\\n-------------\\n\\n`LoRA `_ is an adapter-based method for\\nparameter-efficient finetuning that adds trainable low-rank decomposition matrices to different layers of a neural network,\\nthen freezes the network's remaining parameters. LoRA is most commonly applied to\\ntransformer models, in which case it is common to add the low-rank matrices\\nto some of the linear projections in each transformer layer's self-attention.\\n\\n.. note::\\n\\n If you're unfamiliar, check out these references for the `definition of rank `_\\n and discussion of `low-rank approximations `_.\\n\\nBy finetuning with LoRA (as opposed to finetuning all model parameters),\\nyou can expect to see memory savings due to a substantial reduction in the\\nnumber of parameters with gradients. When using an optimizer with momentum,\\nlike `AdamW `.\\n.. .. _glossary_fsdp2:\\n\\n\", \"type\": \"text\"}, {\"text\": \"Result 4:\\nDocument_id:cc255\\nContent: 06% of all params are trainable.\\n\\n.. note::\\n If you are directly using the LoRA recipe (as detailed :ref:`here`), you need only pass the\\n relevant checkpoint path. Loading model weights and setting trainable parameters will be taken care\\n of in the recipe.\\n\\n\\n.. _lora_recipe_label:\\n\\nLoRA finetuning recipe in torchtune\\n-----------------------------------\\n\\nFinally, we can put it all together and finetune a model using torchtune's `LoRA recipe `_.\\nMake sure that you have first downloaded the Llama2 weights and tokenizer by following :ref:`these instructions`.\\nYou can then run the following command to perform a LoRA finetune of Llama2-7B with two GPUs (each having VRAM of at least 16GB):\\n\\n.. code-block:: bash\\n\\n tune run --nnodes 1 --nproc_per_node 2 lora_finetune_distributed --config llama2/7B_lora\\n\\n.. note::\\n Make sure to point to the location of your Llama2 weights and tokenizer. This can be done\\n either by adding :code:`checkpointer.checkpoint_files=[my_model_checkpoint_path] tokenizer_checkpoint=my_tokenizer_checkpoint_path`\\n or by directly modifying the :code:`7B_lora.yaml` file. See our \\\"\\\":ref:`config_tutorial_label`\\\" recipe\\n for more details on how you can easily clone and modify torchtune configs.\\n\\n.. note::\\n You can modify the value of :code:`nproc_per_node` depending on (a) the number of GPUs you have available,\\n and (b) the memory constraints of your hardware.\\n\\nThe preceding command will run a LoRA finetune with torchtune's factory settings, but we may want to experiment a bit.\\nLet's take a closer look at some of the :code:`lora_finetune_distributed` config.\\n\\n.. code-block:: yaml\\n\\n # Model Arguments\\n model:\\n _component_: lora_llama2_7b\\n lora_attn_modules: ['q_proj', 'v_proj']\\n lora_rank: 8\\n lora_alpha: 16\\n ...\\n\\nWe see that the\\n\", \"type\": \"text\"}, {\"text\": \"Result 5:\\nDocument_id:7a06a\\nContent: etune\\n:func:`torchtune.models.llama3.llama3_8b` with DoRA, you would use :func:`torchtune.models.llama3.lora_llama3_8b` with ``use_dora=True``:\\n\\n.. code-block:: bash\\n\\n tune run lora_finetune_single_device --config llama3/8B_lora_single_device \\\\\\n model.use_dora=True\\n\\n.. code-block:: yaml\\n\\n model:\\n _component_: torchtune.models.lora_llama3_8b\\n use_dora: True\\n\\nSince DoRA extends LoRA, the parameters for :ref:`customizing LoRA ` are identical. You can also quantize the base model weights like in :ref:`glossary_qlora` by using ``quantize=True`` to reap\\neven more memory savings!\\n\\n.. code-block:: bash\\n\\n tune run lora_finetune_single_device --config llama3/8B_lora_single_device \\\\\\n model.apply_lora_to_mlp=True \\\\\\n model.lora_attn_modules=[\\\"q_proj\\\",\\\"k_proj\\\",\\\"v_proj\\\"] \\\\\\n model.lora_rank=16 \\\\\\n model.lora_alpha=32 \\\\\\n model.use_dora=True \\\\\\n model.quantize_base=True\\n\\n.. code-block:: yaml\\n\\n model:\\n _component_: torchtune.models.lora_llama3_8b\\n apply_lora_to_mlp: True\\n lora_attn_modules: [\\\"q_proj\\\", \\\"k_proj\\\", \\\"v_proj\\\"]\\n lora_rank: 16\\n lora_alpha: 32\\n use_dora: True\\n quantize_base: True\\n\\n\\n.. note::\\n\\n Under the hood, we've enabled DoRA by adding the :class:`~torchtune.modules.peft.DoRALinear` module, which we swap\\n out for :class:`~torchtune.modules.peft.LoRALinear` when ``use_dora=True``.\\n\\n.. _glossary_distrib:\\n\\n\\n.. TODO\\n\\n.. Distributed\\n.. -----------\\n\\n.. .. _glossary_fsdp:\\n\\n.. Fully Sharded Data Parallel (FSDP)\\n.. ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\\n\\n.. All our ``_distributed`` recipes use `FSDP `.\\n.. .. _glossary_fsdp2:\\n\\n\", \"type\": \"text\"}, {\"text\": \"END of knowledge_search tool results.\\n\", \"type\": \"text\"}], \"role\": \"tool\", \"tool_name\": \"knowledge_search\"}}]], {\"response_format\": null, \"sampling_params\": {\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"SamplingParams\", \"data\": {\"max_tokens\": 0, \"repetition_penalty\": 1.0, \"strategy\": {\"temperature\": 0.0001, \"top_p\": 0.9, \"type\": \"top_p\"}}}, \"stream\": true, \"tool_config\": {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolConfig\", \"data\": {\"system_message_behavior\": {\"__enum__\": \"SystemMessageBehavior\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"append\"}, \"tool_choice\": {\"__enum__\": \"ToolChoice\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"auto\"}, \"tool_prompt_format\": null}}, \"tool_prompt_format\": null, \"tools\": [{\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"ToolDefinition\", \"data\": {\"description\": \"Search for information in a database.\", \"parameters\": {\"query\": {\"default\": null, \"description\": \"The query to search for. Can be a natural language sentence or keywords.\", \"param_type\": \"string\", \"required\": true}}, \"tool_name\": \"knowledge_search\"}}]}]": { + "chunks": [ + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "start" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "I", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "'m ready to help you answer questions about Torchtune based on the", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " documentation you provided. What's your first question?", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "complete" + }, + "logprobs": null, + "stop_reason": { + "__enum__": "StopReason", + "__module__": "llama_stack.models.llama.datatypes", + "value": "end_of_turn" + } + }, + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 75 + }, + { + "metric": "completion_tokens", + "unit": null, + "value": 35 + }, + { + "metric": "total_tokens", + "unit": null, + "value": 110 + } + ] + } } ], "type": "generator" @@ -23085,7 +26713,7 @@ "__module__": "llama_stack.apis.common.content_types", "value": "in_progress" }, - "tool_call": "{\"type\": \"function\", \"name\": \"knowledge_search\", \"parameters", + "tool_call": "{\"type\": \"function\", \"name\": \"knowledge_search\",", "type": "tool_call" }, "event_type": { @@ -23110,7 +26738,7 @@ "__module__": "llama_stack.apis.common.content_types", "value": "in_progress" }, - "tool_call": "\": {\"query\": \"Torchtune documentation\"}}", + "tool_call": " \"parameters\": {\"query\": \"Torchtune documentation\"}}", "type": "tool_call" }, "event_type": { @@ -23139,7 +26767,7 @@ "arguments": { "query": "Torchtune documentation" }, - "call_id": "5cfa4683-2147-41ab-9a44-a8b7f23e9f75", + "call_id": "8b279fd4-0479-48d4-8ef7-882004f994b2", "tool_name": "knowledge_search" }, "type": "tool_call" @@ -23180,7 +26808,23 @@ "value": "end_of_turn" } }, - "metrics": null + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 39 + }, + { + "metric": "completion_tokens", + "unit": null, + "value": 10 + }, + { + "metric": "total_tokens", + "unit": null, + "value": 49 + } + ] } } ], @@ -23234,7 +26878,27 @@ "data": { "event": { "delta": { - "text": " attention type used by Llama3-8B is grouped-query attention.", + "text": " attention type used by Llama3-8B is grouped-query attention", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": ".", "type": "text" }, "event_type": { @@ -23269,7 +26933,23 @@ "value": "end_of_turn" } }, - "metrics": null + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 80 + }, + { + "metric": "completion_tokens", + "unit": null, + "value": 26 + }, + { + "metric": "total_tokens", + "unit": null, + "value": 106 + } + ] } } ], @@ -23323,7 +27003,7 @@ "data": { "event": { "delta": { - "text": " attention type used by Llama3", + "text": " attention type used by Llama3-8B is grouped", "type": "text" }, "event_type": { @@ -23343,7 +27023,7 @@ "data": { "event": { "delta": { - "text": "-8B is grouped-query attention.", + "text": "-query attention.", "type": "text" }, "event_type": { @@ -23378,7 +27058,23 @@ "value": "end_of_turn" } }, - "metrics": null + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 80 + }, + { + "metric": "completion_tokens", + "unit": null, + "value": 26 + }, + { + "metric": "total_tokens", + "unit": null, + "value": 106 + } + ] } } ], @@ -23432,7 +27128,7 @@ "data": { "event": { "delta": { - "text": " \"type\": \"function\",\n ", + "text": " \"type\": \"function\",\n \"name\": \"knowledge", "type": "text" }, "event_type": { @@ -23452,27 +27148,7 @@ "data": { "event": { "delta": { - "text": " \"name\": \"knowledge_search\",\n ", - "type": "text" - }, - "event_type": { - "__enum__": "ChatCompletionResponseEventType", - "__module__": "llama_stack.apis.inference.inference", - "value": "progress" - }, - "logprobs": null, - "stop_reason": null - }, - "metrics": null - } - }, - { - "__module__": "llama_stack.apis.inference.inference", - "__pydantic__": "ChatCompletionResponseStreamChunk", - "data": { - "event": { - "delta": { - "text": " \"parameters\": {\n \"query", + "text": "_search\",\n \"parameters\": {\n \"query", "type": "text" }, "event_type": { @@ -23521,7 +27197,7 @@ "arguments": { "query": "Llama3-8B attention type" }, - "call_id": "b2d62231-df92-43ed-b51f-f7b8a4bc4b15", + "call_id": "34e18a7e-5c52-403b-9dd3-cdb3f7f8cb89", "tool_name": "knowledge_search" }, "type": "tool_call" @@ -23562,7 +27238,23 @@ "value": "end_of_turn" } }, - "metrics": null + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 40 + }, + { + "metric": "completion_tokens", + "unit": null, + "value": 48 + }, + { + "metric": "total_tokens", + "unit": null, + "value": 88 + } + ] } } ], @@ -23626,7 +27318,7 @@ "__module__": "llama_stack.apis.common.content_types", "value": "in_progress" }, - "tool_call": "{\"type\": \"function\", \"name\":", + "tool_call": "{\"type\": \"function\", \"name\": \"knowledge_search\", \"parameters", "type": "tool_call" }, "event_type": { @@ -23651,32 +27343,7 @@ "__module__": "llama_stack.apis.common.content_types", "value": "in_progress" }, - "tool_call": " \"knowledge_search\", \"parameters\": {\"query\": \"Llama", - "type": "tool_call" - }, - "event_type": { - "__enum__": "ChatCompletionResponseEventType", - "__module__": "llama_stack.apis.inference.inference", - "value": "progress" - }, - "logprobs": null, - "stop_reason": null - }, - "metrics": null - } - }, - { - "__module__": "llama_stack.apis.inference.inference", - "__pydantic__": "ChatCompletionResponseStreamChunk", - "data": { - "event": { - "delta": { - "parse_status": { - "__enum__": "ToolCallParseStatus", - "__module__": "llama_stack.apis.common.content_types", - "value": "in_progress" - }, - "tool_call": "3-8B attention type\"}}", + "tool_call": "\": {\"query\": \"Llama3-8B attention type\"}}", "type": "tool_call" }, "event_type": { @@ -23705,7 +27372,7 @@ "arguments": { "query": "Llama3-8B attention type" }, - "call_id": "52c2b1ea-3695-4030-87a1-d0ca6d1056af", + "call_id": "2c25de18-5466-4f97-bef1-b463abbbcd55", "tool_name": "knowledge_search" }, "type": "tool_call" @@ -23746,7 +27413,23 @@ "value": "end_of_turn" } }, - "metrics": null + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 40 + }, + { + "metric": "completion_tokens", + "unit": null, + "value": 10 + }, + { + "metric": "total_tokens", + "unit": null, + "value": 50 + } + ] } } ], @@ -24269,7 +27952,596 @@ ], "type": "generator" }, + "[[\"meta-llama/Llama-3.1-8B-Instruct\", [{\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"SystemMessage\", \"data\": {\"content\": \"You are a helpful assistant\", \"role\": \"system\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"UserMessage\", \"data\": {\"content\": \"Search the web and tell me who the founder of Meta is.\", \"context\": null, \"role\": \"user\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"CompletionMessage\", \"data\": {\"content\": \"\", \"role\": \"assistant\", \"stop_reason\": {\"__enum__\": \"StopReason\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"end_of_turn\"}, \"tool_calls\": [{\"arguments\": {\"query\": \"Meta founder\"}, \"call_id\": \"\", \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"brave_search\"}}]}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolResponseMessage\", \"data\": {\"call_id\": \"\", \"content\": \"{\\\"query\\\": \\\"Meta founder\\\", \\\"top_k\\\": [{\\\"title\\\": \\\"Mark Zuckerberg, Founder, Chairman and Chief Executive Officer - Meta\\\", \\\"url\\\": \\\"https://about.meta.com/media-gallery/executives/mark-zuckerberg/\\\", \\\"content\\\": \\\"Mark Zuckerberg, Founder, Chairman and Chief Executive Officer | Meta Meta Quest Ray-Ban Meta Meta Horizon Meta AI Meta Verified Meta Pay Meta Horizon Workrooms Meta and you Learn about our community Shop Meta Meta Quest Meta Portal Meta Horizon Mark Zuckerberg is the founder, chairman and CEO of Meta, which he originally founded as Facebook in 2004. In October 2021, Facebook rebranded to Meta to reflect all of its products and services across its family of apps and a focus on developing social experiences for the metaverse \\\\u2014 moving beyond 2D screens toward immersive experiences like augmented and virtual reality to help build the next evolution in social technology. Shop Ray-Ban Meta glassesRay-Ban StoriesPrivacy informationSupported countries \\\\u00a9 2025 Meta\\\", \\\"score\\\": 0.81595254, \\\"raw_content\\\": null}, {\\\"title\\\": \\\"Executives - Meta\\\", \\\"url\\\": \\\"https://about.meta.com/media-gallery/executives/\\\", \\\"content\\\": \\\"Mark Zuckerberg, Founder, Chairman and Chief Executive Officer Joel Kaplan, Chief Global Affairs Officer Susan Li, Chief Financial Officer Javier Olivan, Chief Operating Officer Chris Cox, Chief Product Officer Andrew \\\\u2018Boz\\\\u2019 Bosworth, Chief Technology Officer Jennifer Newstead, Chief Legal Officer Dave Wehner, Chief Strategy Officer Will Cathcart, Head of WhatsApp Naomi Gleit, Head of Product John Hegeman, Chief Revenue Officer Adam Mosseri, Head of Instagram Erin Egan, Chief Privacy Officer, Policy Michel Protti, Chief Privacy Officer, Product Alex Schultz, Chief Marketing Officer and VP of Analytics Tom Alison, Head of Facebook Nicola Mendelsohn, Head of Global Business Group Ahmad Al-Dahle, VP and Head of GenAI at Meta Joelle Pineau, Vice President of AI Research and Head of FAIR at Meta\\\", \\\"score\\\": 0.70726365, \\\"raw_content\\\": null}, {\\\"title\\\": \\\"Meta - Leadership & Governance\\\", \\\"url\\\": \\\"https://investor.atmeta.com/leadership-and-governance/\\\", \\\"content\\\": \\\"Mr. Andreessen was a co-founder of Netscape Communications Corporation, a software company, serving in various positions, including Chief Technology Officer and Executive Vice President of Products. Ms. Killefer also served as Assistant Secretary for Management, Chief Financial Officer, and Chief Operating Officer of the U.S. Department of the Treasury from 1997 to 2000 and as a member of the IRS Oversight Board from 2000 to 2005, including as Chair of the IRS Oversight Board from 2002 to 2004. Ms. Travis has served as Executive Vice President and Chief Financial Officer of The Estee Lauder Companies Inc., a global manufacturer and marketer of skin care, makeup, fragrance and hair care products, since August 2012.\\\", \\\"score\\\": 0.467308, \\\"raw_content\\\": null}, {\\\"title\\\": \\\"Meta Platforms - Wikipedia\\\", \\\"url\\\": \\\"https://en.wikipedia.org/wiki/Meta_Platforms\\\", \\\"content\\\": \\\"Following a period of intense scrutiny and damaging whistleblower leaks, news started to emerge on October 21, 2021, about Facebook's plan to rebrand the company and change its name.[15][54] In the Q3 2021 Earnings Call on October 25, Mark Zuckerberg discussed the ongoing criticism of the company's social services and the way it operates, and pointed to the pivoting efforts to building the metaverse \\\\u2013 without mentioning the rebranding and the name change.[55] The metaverse vision and the name change from Facebook, Inc. to Meta Platforms was introduced at Facebook Connect on October 28, 2021.[16] Based on Facebook's PR campaign, the name change reflects the company's shifting long term focus of building the metaverse, a digital extension of the physical world by social media, virtual reality and augmented reality features.[16][56]\\\", \\\"score\\\": 0.14999175, \\\"raw_content\\\": null}, {\\\"title\\\": \\\"Mark Zuckerberg - Wikipedia\\\", \\\"url\\\": \\\"https://en.wikipedia.org/wiki/Mark_Zuckerberg\\\", \\\"content\\\": \\\"They began dating in 2003.[175] In September 2010, Chan, who was a medical student at the University of California, San Francisco at the time,[176] moved into his rented house in Palo Alto, California.[177][178] They married on May 19, 2012, in the grounds of his mansion in an event that also celebrated her graduation from medical school.[179][180] Zuckerberg revealed in July 2015 that they were expecting a baby girl and that Chan had previously experienced three miscarriages.[181] Their first daughter was born in December 2015.[182] They announced in a Chinese New Year video that their daughter's Chinese name is Chen Mingyu (Chinese: \\\\u9648\\\\u660e\\\\u5b87).[183] Their second daughter was born in August 2017.[184] Zuckerberg and his wife welcomed their third daughter in March 2023 and announced the news across his social media pages.[185] The couple also have a Puli dog named Beast,[186] who has over two million followers on Facebook.[187] Zuckerberg commissioned the visual artist Daniel Arsham to build a 7-foot-tall sculpture of his wife, which was unveiled in 2024.[188]\\\", \\\"score\\\": 0.03678684, \\\"raw_content\\\": null}]}\", \"role\": \"tool\", \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"brave_search\"}}}]], {\"response_format\": null, \"sampling_params\": {\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"SamplingParams\", \"data\": {\"max_tokens\": 0, \"repetition_penalty\": 1.0, \"strategy\": {\"temperature\": 0.0001, \"top_p\": 0.9, \"type\": \"top_p\"}}}, \"stream\": true, \"tool_config\": {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolConfig\", \"data\": {\"system_message_behavior\": {\"__enum__\": \"SystemMessageBehavior\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"append\"}, \"tool_choice\": {\"__enum__\": \"ToolChoice\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"auto\"}, \"tool_prompt_format\": null}}, \"tool_prompt_format\": null, \"tools\": [{\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"ToolDefinition\", \"data\": {\"description\": \"Search the web for information\", \"parameters\": {\"query\": {\"default\": null, \"description\": \"The query to search for\", \"param_type\": \"string\", \"required\": true}}, \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"brave_search\"}}}]}]": { + "chunks": [ + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "start" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "The", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " founder of Meta is Mark Zuckerberg.", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "complete" + }, + "logprobs": null, + "stop_reason": { + "__enum__": "StopReason", + "__module__": "llama_stack.models.llama.datatypes", + "value": "end_of_turn" + } + }, + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 1220 + }, + { + "metric": "completion_tokens", + "unit": null, + "value": 18 + }, + { + "metric": "total_tokens", + "unit": null, + "value": 1238 + } + ] + } + } + ], + "type": "generator" + }, + "[[\"meta-llama/Llama-3.1-8B-Instruct\", [{\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"SystemMessage\", \"data\": {\"content\": \"You are a helpful assistant\", \"role\": \"system\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"UserMessage\", \"data\": {\"content\": \"Search the web and tell me who the founder of Meta is.\", \"context\": null, \"role\": \"user\"}}]], {\"response_format\": null, \"sampling_params\": {\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"SamplingParams\", \"data\": {\"max_tokens\": 0, \"repetition_penalty\": 1.0, \"strategy\": {\"temperature\": 0.0001, \"top_p\": 0.9, \"type\": \"top_p\"}}}, \"stream\": true, \"tool_config\": {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolConfig\", \"data\": {\"system_message_behavior\": {\"__enum__\": \"SystemMessageBehavior\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"append\"}, \"tool_choice\": {\"__enum__\": \"ToolChoice\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"auto\"}, \"tool_prompt_format\": null}}, \"tool_prompt_format\": null, \"tools\": [{\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"ToolDefinition\", \"data\": {\"description\": \"Search the web for information\", \"parameters\": {\"query\": {\"default\": null, \"description\": \"The query to search for\", \"param_type\": \"string\", \"required\": true}}, \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"brave_search\"}}}]}]": { + "chunks": [ + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "start" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "parse_status": { + "__enum__": "ToolCallParseStatus", + "__module__": "llama_stack.apis.common.content_types", + "value": "started" + }, + "tool_call": "", + "type": "tool_call" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "parse_status": { + "__enum__": "ToolCallParseStatus", + "__module__": "llama_stack.apis.common.content_types", + "value": "in_progress" + }, + "tool_call": "brave_search.call(query=\"Meta founder\")", + "type": "tool_call" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "parse_status": { + "__enum__": "ToolCallParseStatus", + "__module__": "llama_stack.apis.common.content_types", + "value": "succeeded" + }, + "tool_call": { + "arguments": { + "query": "Meta founder" + }, + "call_id": "3918bbe3-28b2-411d-be4a-f0eb2aa137e8", + "tool_name": { + "__enum__": "BuiltinTool", + "__module__": "llama_stack.models.llama.datatypes", + "value": "brave_search" + } + }, + "type": "tool_call" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": { + "__enum__": "StopReason", + "__module__": "llama_stack.models.llama.datatypes", + "value": "end_of_turn" + } + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "complete" + }, + "logprobs": null, + "stop_reason": { + "__enum__": "StopReason", + "__module__": "llama_stack.models.llama.datatypes", + "value": "end_of_turn" + } + }, + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 33 + }, + { + "metric": "completion_tokens", + "unit": null, + "value": 10 + }, + { + "metric": "total_tokens", + "unit": null, + "value": 43 + } + ] + } + } + ], + "type": "generator" + }, "[[\"meta-llama/Llama-3.1-8B-Instruct\", [{\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"SystemMessage\", \"data\": {\"content\": \"You are a helpful assistant\", \"role\": \"system\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"UserMessage\", \"data\": {\"content\": \"What is the boiling point of polyjuice?\", \"context\": null, \"role\": \"user\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"CompletionMessage\", \"data\": {\"content\": \"\", \"role\": \"assistant\", \"stop_reason\": {\"__enum__\": \"StopReason\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"end_of_turn\"}, \"tool_calls\": [{\"arguments\": {\"liquid_name\": \"polyjuice\"}, \"call_id\": \"\", \"tool_name\": \"get_boiling_point\"}]}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolResponseMessage\", \"data\": {\"call_id\": \"\", \"content\": \"-100\", \"role\": \"tool\", \"tool_name\": \"get_boiling_point\"}}]], {\"response_format\": null, \"sampling_params\": {\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"SamplingParams\", \"data\": {\"max_tokens\": 0, \"repetition_penalty\": 1.0, \"strategy\": {\"temperature\": 0.0001, \"top_p\": 0.9, \"type\": \"top_p\"}}}, \"stream\": true, \"tool_config\": {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolConfig\", \"data\": {\"system_message_behavior\": {\"__enum__\": \"SystemMessageBehavior\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"append\"}, \"tool_choice\": \"get_boiling_point\", \"tool_prompt_format\": null}}, \"tool_prompt_format\": null, \"tools\": [{\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"ToolDefinition\", \"data\": {\"description\": \"Returns the boiling point of a liquid in Celcius or Fahrenheit\", \"parameters\": {\"celcius\": {\"default\": true, \"description\": \"Whether to return the boiling point in Celcius\", \"param_type\": \"bool\", \"required\": false}, \"liquid_name\": {\"default\": null, \"description\": \"The name of the liquid\", \"param_type\": \"string\", \"required\": true}}, \"tool_name\": \"get_boiling_point\"}}]}]": { + "chunks": [ + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "start" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "The", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " function `get_boiling_point` is", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " not able to find the boiling point of poly", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "juice as it is a fictional liquid from the Harry Potter series", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": ". The function is only able to find the boiling point of real", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " liquids.", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "complete" + }, + "logprobs": null, + "stop_reason": { + "__enum__": "StopReason", + "__module__": "llama_stack.models.llama.datatypes", + "value": "end_of_turn" + } + }, + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 70 + }, + { + "metric": "completion_tokens", + "unit": null, + "value": 56 + }, + { + "metric": "total_tokens", + "unit": null, + "value": 126 + } + ] + } + } + ], + "type": "generator" + }, + "[[\"meta-llama/Llama-3.1-8B-Instruct\", [{\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"SystemMessage\", \"data\": {\"content\": \"You are a helpful assistant\", \"role\": \"system\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"UserMessage\", \"data\": {\"content\": \"What is the boiling point of polyjuice?\", \"context\": null, \"role\": \"user\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"CompletionMessage\", \"data\": {\"content\": \"\", \"role\": \"assistant\", \"stop_reason\": {\"__enum__\": \"StopReason\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"end_of_turn\"}, \"tool_calls\": [{\"arguments\": {\"liquid_name\": \"polyjuice\"}, \"call_id\": \"\", \"tool_name\": \"get_boiling_point\"}]}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolResponseMessage\", \"data\": {\"call_id\": \"\", \"content\": \"-100\", \"role\": \"tool\", \"tool_name\": \"get_boiling_point\"}}]], {\"response_format\": null, \"sampling_params\": {\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"SamplingParams\", \"data\": {\"max_tokens\": 0, \"repetition_penalty\": 1.0, \"strategy\": {\"temperature\": 0.0001, \"top_p\": 0.9, \"type\": \"top_p\"}}}, \"stream\": true, \"tool_config\": {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolConfig\", \"data\": {\"system_message_behavior\": {\"__enum__\": \"SystemMessageBehavior\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"append\"}, \"tool_choice\": {\"__enum__\": \"ToolChoice\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"auto\"}, \"tool_prompt_format\": null}}, \"tool_prompt_format\": null, \"tools\": [{\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"ToolDefinition\", \"data\": {\"description\": \"Returns the boiling point of a liquid in Celcius or Fahrenheit\", \"parameters\": {\"celcius\": {\"default\": true, \"description\": \"Whether to return the boiling point in Celcius\", \"param_type\": \"bool\", \"required\": false}, \"liquid_name\": {\"default\": null, \"description\": \"The name of the liquid\", \"param_type\": \"string\", \"required\": true}}, \"tool_name\": \"get_boiling_point\"}}, {\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"ToolDefinition\", \"data\": {\"description\": \"Search the web for information\", \"parameters\": {\"query\": {\"default\": null, \"description\": \"The query to search for\", \"param_type\": \"string\", \"required\": true}}, \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"brave_search\"}}}]}]": { + "chunks": [ + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "start" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "The", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " function `get_boiling_point", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "` is not able to find the boiling point of polyjuice", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " as it is not a real liquid.", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "complete" + }, + "logprobs": null, + "stop_reason": { + "__enum__": "StopReason", + "__module__": "llama_stack.models.llama.datatypes", + "value": "end_of_turn" + } + }, + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 70 + }, + { + "metric": "completion_tokens", + "unit": null, + "value": 38 + }, + { + "metric": "total_tokens", + "unit": null, + "value": 108 + } + ] + } + } + ], + "type": "generator" + }, + "[[\"meta-llama/Llama-3.1-8B-Instruct\", [{\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"SystemMessage\", \"data\": {\"content\": \"You are a helpful assistant\", \"role\": \"system\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"UserMessage\", \"data\": {\"content\": \"What is the boiling point of polyjuice?\", \"context\": null, \"role\": \"user\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"CompletionMessage\", \"data\": {\"content\": \"\", \"role\": \"assistant\", \"stop_reason\": {\"__enum__\": \"StopReason\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"end_of_turn\"}, \"tool_calls\": [{\"arguments\": {\"liquid_name\": \"polyjuice\"}, \"call_id\": \"\", \"tool_name\": \"get_boiling_point\"}]}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolResponseMessage\", \"data\": {\"call_id\": \"\", \"content\": \"-100\", \"role\": \"tool\", \"tool_name\": \"get_boiling_point\"}}]], {\"response_format\": null, \"sampling_params\": {\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"SamplingParams\", \"data\": {\"max_tokens\": 0, \"repetition_penalty\": 1.0, \"strategy\": {\"temperature\": 0.0001, \"top_p\": 0.9, \"type\": \"top_p\"}}}, \"stream\": true, \"tool_config\": {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolConfig\", \"data\": {\"system_message_behavior\": {\"__enum__\": \"SystemMessageBehavior\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"append\"}, \"tool_choice\": {\"__enum__\": \"ToolChoice\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"required\"}, \"tool_prompt_format\": null}}, \"tool_prompt_format\": null, \"tools\": [{\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"ToolDefinition\", \"data\": {\"description\": \"Returns the boiling point of a liquid in Celcius or Fahrenheit\", \"parameters\": {\"celcius\": {\"default\": true, \"description\": \"Whether to return the boiling point in Celcius\", \"param_type\": \"bool\", \"required\": false}, \"liquid_name\": {\"default\": null, \"description\": \"The name of the liquid\", \"param_type\": \"string\", \"required\": true}}, \"tool_name\": \"get_boiling_point\"}}]}]": { "chunks": [ { "__module__": "llama_stack.apis.inference.inference", @@ -24337,7 +28609,7 @@ "data": { "event": { "delta": { - "text": " find the boiling point of polyjuice as", + "text": " find the boiling point of polyju", "type": "text" }, "event_type": { @@ -24357,27 +28629,7 @@ "data": { "event": { "delta": { - "text": " it is a fictional liquid from the Harry Potter series. The", - "type": "text" - }, - "event_type": { - "__enum__": "ChatCompletionResponseEventType", - "__module__": "llama_stack.apis.inference.inference", - "value": "progress" - }, - "logprobs": null, - "stop_reason": null - }, - "metrics": null - } - }, - { - "__module__": "llama_stack.apis.inference.inference", - "__pydantic__": "ChatCompletionResponseStreamChunk", - "data": { - "event": { - "delta": { - "text": " function is only able to find the boiling point of real liquids.", + "text": "ice as it is not a real liquid.", "type": "text" }, "event_type": { @@ -24412,265 +28664,23 @@ "value": "end_of_turn" } }, - "metrics": null - } - } - ], - "type": "generator" - }, - "[[\"meta-llama/Llama-3.1-8B-Instruct\", [{\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"SystemMessage\", \"data\": {\"content\": \"You are a helpful assistant\", \"role\": \"system\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"UserMessage\", \"data\": {\"content\": \"What is the boiling point of polyjuice?\", \"context\": null, \"role\": \"user\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"CompletionMessage\", \"data\": {\"content\": \"\", \"role\": \"assistant\", \"stop_reason\": {\"__enum__\": \"StopReason\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"end_of_turn\"}, \"tool_calls\": [{\"arguments\": {\"liquid_name\": \"polyjuice\"}, \"call_id\": \"\", \"tool_name\": \"get_boiling_point\"}]}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolResponseMessage\", \"data\": {\"call_id\": \"\", \"content\": \"-100\", \"role\": \"tool\", \"tool_name\": \"get_boiling_point\"}}]], {\"response_format\": null, \"sampling_params\": {\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"SamplingParams\", \"data\": {\"max_tokens\": 0, \"repetition_penalty\": 1.0, \"strategy\": {\"temperature\": 0.0001, \"top_p\": 0.9, \"type\": \"top_p\"}}}, \"stream\": true, \"tool_config\": {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolConfig\", \"data\": {\"system_message_behavior\": {\"__enum__\": \"SystemMessageBehavior\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"append\"}, \"tool_choice\": {\"__enum__\": \"ToolChoice\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"auto\"}, \"tool_prompt_format\": null}}, \"tool_prompt_format\": null, \"tools\": [{\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"ToolDefinition\", \"data\": {\"description\": \"Returns the boiling point of a liquid in Celcius or Fahrenheit\", \"parameters\": {\"celcius\": {\"default\": true, \"description\": \"Whether to return the boiling point in Celcius\", \"param_type\": \"bool\", \"required\": false}, \"liquid_name\": {\"default\": null, \"description\": \"The name of the liquid\", \"param_type\": \"string\", \"required\": true}}, \"tool_name\": \"get_boiling_point\"}}, {\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"ToolDefinition\", \"data\": {\"description\": \"Search the web for information\", \"parameters\": {\"query\": {\"default\": null, \"description\": \"The query to search for\", \"param_type\": \"string\", \"required\": true}}, \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"brave_search\"}}}]}]": { - "chunks": [ - { - "__module__": "llama_stack.apis.inference.inference", - "__pydantic__": "ChatCompletionResponseStreamChunk", - "data": { - "event": { - "delta": { - "text": "", - "type": "text" + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 70 }, - "event_type": { - "__enum__": "ChatCompletionResponseEventType", - "__module__": "llama_stack.apis.inference.inference", - "value": "start" + { + "metric": "completion_tokens", + "unit": null, + "value": 38 }, - "logprobs": null, - "stop_reason": null - }, - "metrics": null - } - }, - { - "__module__": "llama_stack.apis.inference.inference", - "__pydantic__": "ChatCompletionResponseStreamChunk", - "data": { - "event": { - "delta": { - "text": "The", - "type": "text" - }, - "event_type": { - "__enum__": "ChatCompletionResponseEventType", - "__module__": "llama_stack.apis.inference.inference", - "value": "progress" - }, - "logprobs": null, - "stop_reason": null - }, - "metrics": null - } - }, - { - "__module__": "llama_stack.apis.inference.inference", - "__pydantic__": "ChatCompletionResponseStreamChunk", - "data": { - "event": { - "delta": { - "text": " function `get_boiling_point` is not able to find the", - "type": "text" - }, - "event_type": { - "__enum__": "ChatCompletionResponseEventType", - "__module__": "llama_stack.apis.inference.inference", - "value": "progress" - }, - "logprobs": null, - "stop_reason": null - }, - "metrics": null - } - }, - { - "__module__": "llama_stack.apis.inference.inference", - "__pydantic__": "ChatCompletionResponseStreamChunk", - "data": { - "event": { - "delta": { - "text": " boiling point of polyjuice as it is not a real liquid", - "type": "text" - }, - "event_type": { - "__enum__": "ChatCompletionResponseEventType", - "__module__": "llama_stack.apis.inference.inference", - "value": "progress" - }, - "logprobs": null, - "stop_reason": null - }, - "metrics": null - } - }, - { - "__module__": "llama_stack.apis.inference.inference", - "__pydantic__": "ChatCompletionResponseStreamChunk", - "data": { - "event": { - "delta": { - "text": ".", - "type": "text" - }, - "event_type": { - "__enum__": "ChatCompletionResponseEventType", - "__module__": "llama_stack.apis.inference.inference", - "value": "progress" - }, - "logprobs": null, - "stop_reason": null - }, - "metrics": null - } - }, - { - "__module__": "llama_stack.apis.inference.inference", - "__pydantic__": "ChatCompletionResponseStreamChunk", - "data": { - "event": { - "delta": { - "text": "", - "type": "text" - }, - "event_type": { - "__enum__": "ChatCompletionResponseEventType", - "__module__": "llama_stack.apis.inference.inference", - "value": "complete" - }, - "logprobs": null, - "stop_reason": { - "__enum__": "StopReason", - "__module__": "llama_stack.models.llama.datatypes", - "value": "end_of_turn" + { + "metric": "total_tokens", + "unit": null, + "value": 108 } - }, - "metrics": null - } - } - ], - "type": "generator" - }, - "[[\"meta-llama/Llama-3.1-8B-Instruct\", [{\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"SystemMessage\", \"data\": {\"content\": \"You are a helpful assistant\", \"role\": \"system\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"UserMessage\", \"data\": {\"content\": \"What is the boiling point of polyjuice?\", \"context\": null, \"role\": \"user\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"CompletionMessage\", \"data\": {\"content\": \"\", \"role\": \"assistant\", \"stop_reason\": {\"__enum__\": \"StopReason\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"end_of_turn\"}, \"tool_calls\": [{\"arguments\": {\"liquid_name\": \"polyjuice\"}, \"call_id\": \"\", \"tool_name\": \"get_boiling_point\"}]}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolResponseMessage\", \"data\": {\"call_id\": \"\", \"content\": \"-100\", \"role\": \"tool\", \"tool_name\": \"get_boiling_point\"}}]], {\"response_format\": null, \"sampling_params\": {\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"SamplingParams\", \"data\": {\"max_tokens\": 0, \"repetition_penalty\": 1.0, \"strategy\": {\"temperature\": 0.0001, \"top_p\": 0.9, \"type\": \"top_p\"}}}, \"stream\": true, \"tool_config\": {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolConfig\", \"data\": {\"system_message_behavior\": {\"__enum__\": \"SystemMessageBehavior\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"append\"}, \"tool_choice\": {\"__enum__\": \"ToolChoice\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"required\"}, \"tool_prompt_format\": null}}, \"tool_prompt_format\": null, \"tools\": [{\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"ToolDefinition\", \"data\": {\"description\": \"Returns the boiling point of a liquid in Celcius or Fahrenheit\", \"parameters\": {\"celcius\": {\"default\": true, \"description\": \"Whether to return the boiling point in Celcius\", \"param_type\": \"bool\", \"required\": false}, \"liquid_name\": {\"default\": null, \"description\": \"The name of the liquid\", \"param_type\": \"string\", \"required\": true}}, \"tool_name\": \"get_boiling_point\"}}]}]": { - "chunks": [ - { - "__module__": "llama_stack.apis.inference.inference", - "__pydantic__": "ChatCompletionResponseStreamChunk", - "data": { - "event": { - "delta": { - "text": "", - "type": "text" - }, - "event_type": { - "__enum__": "ChatCompletionResponseEventType", - "__module__": "llama_stack.apis.inference.inference", - "value": "start" - }, - "logprobs": null, - "stop_reason": null - }, - "metrics": null - } - }, - { - "__module__": "llama_stack.apis.inference.inference", - "__pydantic__": "ChatCompletionResponseStreamChunk", - "data": { - "event": { - "delta": { - "text": "The", - "type": "text" - }, - "event_type": { - "__enum__": "ChatCompletionResponseEventType", - "__module__": "llama_stack.apis.inference.inference", - "value": "progress" - }, - "logprobs": null, - "stop_reason": null - }, - "metrics": null - } - }, - { - "__module__": "llama_stack.apis.inference.inference", - "__pydantic__": "ChatCompletionResponseStreamChunk", - "data": { - "event": { - "delta": { - "text": " function `get_boiling_point` is not able to find the", - "type": "text" - }, - "event_type": { - "__enum__": "ChatCompletionResponseEventType", - "__module__": "llama_stack.apis.inference.inference", - "value": "progress" - }, - "logprobs": null, - "stop_reason": null - }, - "metrics": null - } - }, - { - "__module__": "llama_stack.apis.inference.inference", - "__pydantic__": "ChatCompletionResponseStreamChunk", - "data": { - "event": { - "delta": { - "text": " boiling point of polyjuice as it is not a real liquid", - "type": "text" - }, - "event_type": { - "__enum__": "ChatCompletionResponseEventType", - "__module__": "llama_stack.apis.inference.inference", - "value": "progress" - }, - "logprobs": null, - "stop_reason": null - }, - "metrics": null - } - }, - { - "__module__": "llama_stack.apis.inference.inference", - "__pydantic__": "ChatCompletionResponseStreamChunk", - "data": { - "event": { - "delta": { - "text": ".", - "type": "text" - }, - "event_type": { - "__enum__": "ChatCompletionResponseEventType", - "__module__": "llama_stack.apis.inference.inference", - "value": "progress" - }, - "logprobs": null, - "stop_reason": null - }, - "metrics": null - } - }, - { - "__module__": "llama_stack.apis.inference.inference", - "__pydantic__": "ChatCompletionResponseStreamChunk", - "data": { - "event": { - "delta": { - "text": "", - "type": "text" - }, - "event_type": { - "__enum__": "ChatCompletionResponseEventType", - "__module__": "llama_stack.apis.inference.inference", - "value": "complete" - }, - "logprobs": null, - "stop_reason": { - "__enum__": "StopReason", - "__module__": "llama_stack.models.llama.datatypes", - "value": "end_of_turn" - } - }, - "metrics": null + ] } } ], @@ -24734,7 +28744,7 @@ "__module__": "llama_stack.apis.common.content_types", "value": "in_progress" }, - "tool_call": "{\"type\": \"function\", \"name\": \"get_boiling", + "tool_call": "{\"type\": \"function\", \"name\": \"", "type": "tool_call" }, "event_type": { @@ -24759,7 +28769,7 @@ "__module__": "llama_stack.apis.common.content_types", "value": "in_progress" }, - "tool_call": "_point\", \"parameters\": {\"liquid_name\": \"poly", + "tool_call": "get_boiling_point\", \"parameters\": {\"liquid_name\": \"", "type": "tool_call" }, "event_type": { @@ -24784,7 +28794,7 @@ "__module__": "llama_stack.apis.common.content_types", "value": "in_progress" }, - "tool_call": "juice\"}}", + "tool_call": "polyjuice\"}}", "type": "tool_call" }, "event_type": { @@ -24813,7 +28823,7 @@ "arguments": { "liquid_name": "polyjuice" }, - "call_id": "83d9f330-4c7a-4dd3-8fcb-ccc5301c1f83", + "call_id": "98ea49c2-7788-4fda-8d84-4a584eb1dd27", "tool_name": "get_boiling_point" }, "type": "tool_call" @@ -24854,7 +28864,23 @@ "value": "end_of_turn" } }, - "metrics": null + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 30 + }, + { + "metric": "completion_tokens", + "unit": null, + "value": 10 + }, + { + "metric": "total_tokens", + "unit": null, + "value": 40 + } + ] } } ], @@ -24918,7 +28944,7 @@ "__module__": "llama_stack.apis.common.content_types", "value": "in_progress" }, - "tool_call": "{\"type\": \"function\", \"name\":", + "tool_call": "{\"type\": \"function\", \"name\": \"get", "type": "tool_call" }, "event_type": { @@ -24943,7 +28969,7 @@ "__module__": "llama_stack.apis.common.content_types", "value": "in_progress" }, - "tool_call": " \"get_boiling_point\", \"parameters\": {\"liquid_name\":", + "tool_call": "_boiling_point\", \"parameters\": {\"liquid_name\": \"polyju", "type": "tool_call" }, "event_type": { @@ -24968,7 +28994,7 @@ "__module__": "llama_stack.apis.common.content_types", "value": "in_progress" }, - "tool_call": " \"polyjuice\"}}", + "tool_call": "ice\"}}", "type": "tool_call" }, "event_type": { @@ -24997,7 +29023,7 @@ "arguments": { "liquid_name": "polyjuice" }, - "call_id": "98c63572-06c8-4cc0-a14e-3b10fb9ddc19", + "call_id": "a9d6416d-a739-49b0-b7df-63694412a376", "tool_name": "get_boiling_point" }, "type": "tool_call" @@ -25038,7 +29064,23 @@ "value": "end_of_turn" } }, - "metrics": null + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 30 + }, + { + "metric": "completion_tokens", + "unit": null, + "value": 10 + }, + { + "metric": "total_tokens", + "unit": null, + "value": 40 + } + ] } } ], @@ -25092,7 +29134,7 @@ "data": { "event": { "delta": { - "text": " couldn't find any information on the boiling point of Polyjuice", + "text": " couldn't find any information on the boiling point of", "type": "text" }, "event_type": { @@ -25112,7 +29154,7 @@ "data": { "event": { "delta": { - "text": ". Polyjuice is a magical potion in the Harry Potter series", + "text": " Polyjuice. Polyjuice is a magical potion in the", "type": "text" }, "event_type": { @@ -25132,7 +29174,7 @@ "data": { "event": { "delta": { - "text": " that allows the drinker to transform into someone else. It's", + "text": " Harry Potter series that allows the drinker to transform into someone else", "type": "text" }, "event_type": { @@ -25152,7 +29194,7 @@ "data": { "event": { "delta": { - "text": " not a physical substance with a boiling point. If you have any", + "text": ". It's not a physical substance with a boiling point. If", "type": "text" }, "event_type": { @@ -25172,7 +29214,27 @@ "data": { "event": { "delta": { - "text": " other questions, I'd be happy to help.", + "text": " you have any other questions, I'd be happy to", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " help.", "type": "text" }, "event_type": { @@ -25207,7 +29269,23 @@ "value": "end_of_turn" } }, - "metrics": null + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 30 + }, + { + "metric": "completion_tokens", + "unit": null, + "value": 73 + }, + { + "metric": "total_tokens", + "unit": null, + "value": 103 + } + ] } } ], @@ -25271,7 +29349,7 @@ "__module__": "llama_stack.apis.common.content_types", "value": "in_progress" }, - "tool_call": "{\"type\": \"function\", \"name\":", + "tool_call": "{\"type\": \"function\", \"name\": \"", "type": "tool_call" }, "event_type": { @@ -25296,7 +29374,7 @@ "__module__": "llama_stack.apis.common.content_types", "value": "in_progress" }, - "tool_call": " \"get_boiling_point\", \"parameters\": {\"liquid_name\":", + "tool_call": "get_boiling_point\", \"parameters\": {\"liquid_name\":", "type": "tool_call" }, "event_type": { @@ -25350,7 +29428,7 @@ "arguments": { "liquid_name": "polyjuice" }, - "call_id": "cdccc866-97a0-40fd-b6e2-a0555f0ed921", + "call_id": "722b4201-b9ac-440f-80c4-0e9a7f0b5369", "tool_name": "get_boiling_point" }, "type": "tool_call" @@ -25391,7 +29469,23 @@ "value": "end_of_turn" } }, - "metrics": null + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 30 + }, + { + "metric": "completion_tokens", + "unit": null, + "value": 10 + }, + { + "metric": "total_tokens", + "unit": null, + "value": 40 + } + ] } } ], @@ -25486,6 +29580,111 @@ ], "type": "generator" }, + "[[\"meta-llama/Llama-3.1-8B-Instruct\", [{\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"SystemMessage\", \"data\": {\"content\": \"You are a helpful assistant\", \"role\": \"system\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"UserMessage\", \"data\": {\"content\": \"Write code and execute it to find the answer for: What is the 100th prime number?\", \"context\": null, \"role\": \"user\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"CompletionMessage\", \"data\": {\"content\": \"\", \"role\": \"assistant\", \"stop_reason\": {\"__enum__\": \"StopReason\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"end_of_turn\"}, \"tool_calls\": [{\"arguments\": {\"code\": \"def is_prime(n):\\n if n <= 1:\\n return False\\n if n <= 3:\\n return True\\n if n % 2 == 0 or n % 3 == 0:\\n return False\\n i = 5\\n while i * i <= n:\\n if n % i == 0 or n % (i + 2) == 0:\\n return False\\n i += 6\\n return True\\n\\ndef get_nth_prime(n):\\n count = 0\\n num = 2\\n while True:\\n if is_prime(num):\\n count += 1\\n if count == n:\\n return num\\n num += 1\\n\\nprint(get_nth_prime(100))\"}, \"call_id\": \"\", \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}]}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolResponseMessage\", \"data\": {\"call_id\": \"\", \"content\": \"completed\\n[stdout]\\n541\\n[/stdout]\", \"role\": \"tool\", \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}}]], {\"response_format\": null, \"sampling_params\": {\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"SamplingParams\", \"data\": {\"max_tokens\": 0, \"repetition_penalty\": 1.0, \"strategy\": {\"temperature\": 0.0001, \"top_p\": 0.9, \"type\": \"top_p\"}}}, \"stream\": true, \"tool_config\": {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolConfig\", \"data\": {\"system_message_behavior\": {\"__enum__\": \"SystemMessageBehavior\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"append\"}, \"tool_choice\": {\"__enum__\": \"ToolChoice\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"auto\"}, \"tool_prompt_format\": null}}, \"tool_prompt_format\": null, \"tools\": [{\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"ToolDefinition\", \"data\": {\"description\": \"Execute code\", \"parameters\": {\"code\": {\"default\": null, \"description\": \"The code to execute\", \"param_type\": \"string\", \"required\": true}}, \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}}]}]": { + "chunks": [ + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "start" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "The", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": " 100th prime number is 541.", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "text": "", + "type": "text" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "complete" + }, + "logprobs": null, + "stop_reason": { + "__enum__": "StopReason", + "__module__": "llama_stack.models.llama.datatypes", + "value": "end_of_turn" + } + }, + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 217 + }, + { + "metric": "completion_tokens", + "unit": null, + "value": 20 + }, + { + "metric": "total_tokens", + "unit": null, + "value": 237 + } + ] + } + } + ], + "type": "generator" + }, "[[\"meta-llama/Llama-3.1-8B-Instruct\", [{\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"SystemMessage\", \"data\": {\"content\": \"You are a helpful assistant\", \"role\": \"system\"}}, {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"UserMessage\", \"data\": {\"content\": \"Write code and execute it to find the answer for: What is the 100th prime number?\", \"context\": null, \"role\": \"user\"}}]], {\"response_format\": null, \"sampling_params\": {\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"SamplingParams\", \"data\": {\"max_tokens\": 0, \"repetition_penalty\": 1.0, \"strategy\": {\"temperature\": 0.0001, \"top_p\": 0.9, \"type\": \"top_p\"}}}, \"stream\": true, \"tool_config\": {\"__module__\": \"llama_stack.apis.inference.inference\", \"__pydantic__\": \"ToolConfig\", \"data\": {\"system_message_behavior\": {\"__enum__\": \"SystemMessageBehavior\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"append\"}, \"tool_choice\": {\"__enum__\": \"ToolChoice\", \"__module__\": \"llama_stack.apis.inference.inference\", \"value\": \"auto\"}, \"tool_prompt_format\": null}}, \"tool_prompt_format\": null, \"tools\": [{\"__module__\": \"llama_stack.models.llama.datatypes\", \"__pydantic__\": \"ToolDefinition\", \"data\": {\"description\": \"Execute code\", \"parameters\": {\"code\": {\"default\": null, \"description\": \"The code to execute\", \"param_type\": \"string\", \"required\": true}}, \"tool_name\": {\"__enum__\": \"BuiltinTool\", \"__module__\": \"llama_stack.models.llama.datatypes\", \"value\": \"code_interpreter\"}}}]}]": { "chunks": [ { @@ -25544,7 +29743,7 @@ "__module__": "llama_stack.apis.common.content_types", "value": "in_progress" }, - "tool_call": "def is_prime(n):\n if n <= 1:\n ", + "tool_call": "def is_prime(n):\n if n <= 1:\n return False\n", "type": "tool_call" }, "event_type": { @@ -25569,7 +29768,7 @@ "__module__": "llama_stack.apis.common.content_types", "value": "in_progress" }, - "tool_call": " return False\n if n <= 3:\n return True", + "tool_call": " if n <= 3:\n return True\n if n % ", "type": "tool_call" }, "event_type": { @@ -25594,7 +29793,7 @@ "__module__": "llama_stack.apis.common.content_types", "value": "in_progress" }, - "tool_call": "\n if n % 2 == 0 or n % 3 == 0:\n return False\n i = 5\n while i * i <= n:\n if n % i == 0 or n % (i +", + "tool_call": "2 == 0 or n % 3 == 0:\n return False", "type": "tool_call" }, "event_type": { @@ -25619,7 +29818,7 @@ "__module__": "llama_stack.apis.common.content_types", "value": "in_progress" }, - "tool_call": " 2) == 0:\n return False\n ", + "tool_call": "\n i = 5\n while i * i <= n:\n ", "type": "tool_call" }, "event_type": { @@ -25644,7 +29843,7 @@ "__module__": "llama_stack.apis.common.content_types", "value": "in_progress" }, - "tool_call": " i += 6\n return True\n\ndef get_nth_prime(n):\n count", + "tool_call": " if n % i == 0 or n % (i + 2)", "type": "tool_call" }, "event_type": { @@ -25669,7 +29868,7 @@ "__module__": "llama_stack.apis.common.content_types", "value": "in_progress" }, - "tool_call": " = 0\n num = 2\n while True:\n if", + "tool_call": " == 0:\n return False\n i", "type": "tool_call" }, "event_type": { @@ -25694,7 +29893,7 @@ "__module__": "llama_stack.apis.common.content_types", "value": "in_progress" }, - "tool_call": " is_prime(num):\n count += 1", + "tool_call": " += 6\n return True", "type": "tool_call" }, "event_type": { @@ -25719,7 +29918,7 @@ "__module__": "llama_stack.apis.common.content_types", "value": "in_progress" }, - "tool_call": "\n if count == n:\n return num\n num += ", + "tool_call": "\n\n", "type": "tool_call" }, "event_type": { @@ -25744,7 +29943,82 @@ "__module__": "llama_stack.apis.common.content_types", "value": "in_progress" }, - "tool_call": "1\n\nprint(get_nth_prime(100))", + "tool_call": "def get_nth_prime(n):\n count = 0\n num = 2\n while True:\n if is_prime(num):\n ", + "type": "tool_call" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "parse_status": { + "__enum__": "ToolCallParseStatus", + "__module__": "llama_stack.apis.common.content_types", + "value": "in_progress" + }, + "tool_call": " count += 1\n if count == n:\n return", + "type": "tool_call" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "parse_status": { + "__enum__": "ToolCallParseStatus", + "__module__": "llama_stack.apis.common.content_types", + "value": "in_progress" + }, + "tool_call": " num\n num += 1\n\nprint(get_nth_prime(100", + "type": "tool_call" + }, + "event_type": { + "__enum__": "ChatCompletionResponseEventType", + "__module__": "llama_stack.apis.inference.inference", + "value": "progress" + }, + "logprobs": null, + "stop_reason": null + }, + "metrics": null + } + }, + { + "__module__": "llama_stack.apis.inference.inference", + "__pydantic__": "ChatCompletionResponseStreamChunk", + "data": { + "event": { + "delta": { + "parse_status": { + "__enum__": "ToolCallParseStatus", + "__module__": "llama_stack.apis.common.content_types", + "value": "in_progress" + }, + "tool_call": "))", "type": "tool_call" }, "event_type": { @@ -25773,7 +30047,7 @@ "arguments": { "code": "def is_prime(n):\n if n <= 1:\n return False\n if n <= 3:\n return True\n if n % 2 == 0 or n % 3 == 0:\n return False\n i = 5\n while i * i <= n:\n if n % i == 0 or n % (i + 2) == 0:\n return False\n i += 6\n return True\n\ndef get_nth_prime(n):\n count = 0\n num = 2\n while True:\n if is_prime(num):\n count += 1\n if count == n:\n return num\n num += 1\n\nprint(get_nth_prime(100))" }, - "call_id": "7fca0515-82f3-46e1-bbec-eceb8fa5162e", + "call_id": "229f89b4-d07d-4748-b8ae-f805cb52209e", "tool_name": { "__enum__": "BuiltinTool", "__module__": "llama_stack.models.llama.datatypes", @@ -25818,7 +30092,23 @@ "value": "end_of_turn" } }, - "metrics": null + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 40 + }, + { + "metric": "completion_tokens", + "unit": null, + "value": 10 + }, + { + "metric": "total_tokens", + "unit": null, + "value": 50 + } + ] } } ], @@ -25872,7 +30162,7 @@ "data": { "event": { "delta": { - "text": "plexity the company was founded in 2022", + "text": "plexity the company was founded in 202", "type": "text" }, "event_type": { @@ -25892,7 +30182,7 @@ "data": { "event": { "delta": { - "text": ".", + "text": "2.", "type": "text" }, "event_type": { @@ -25927,7 +30217,23 @@ "value": "end_of_turn" } }, - "metrics": null + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 105 + }, + { + "metric": "completion_tokens", + "unit": null, + "value": 22 + }, + { + "metric": "total_tokens", + "unit": null, + "value": 127 + } + ] } } ], @@ -25981,7 +30287,7 @@ "data": { "event": { "delta": { - "text": "type\": \"function\", \"name\": \"knowledge_search\",", + "text": "type\": \"function\", \"name\": \"knowledge_search\", \"", "type": "text" }, "event_type": { @@ -26001,27 +30307,7 @@ "data": { "event": { "delta": { - "text": " \"parameters\": {\"query\": \"Perplexity company founding", - "type": "text" - }, - "event_type": { - "__enum__": "ChatCompletionResponseEventType", - "__module__": "llama_stack.apis.inference.inference", - "value": "progress" - }, - "logprobs": null, - "stop_reason": null - }, - "metrics": null - } - }, - { - "__module__": "llama_stack.apis.inference.inference", - "__pydantic__": "ChatCompletionResponseStreamChunk", - "data": { - "event": { - "delta": { - "text": " date\"}}", + "text": "parameters\": {\"query\": \"Perplexity company founding date\"}}", "type": "text" }, "event_type": { @@ -26050,7 +30336,7 @@ "arguments": { "query": "Perplexity company founding date" }, - "call_id": "ca248109-25af-4737-90cb-6461faaf4e63", + "call_id": "5ea88dde-f090-4157-9219-45a16100ef21", "tool_name": "knowledge_search" }, "type": "tool_call" @@ -26091,7 +30377,23 @@ "value": "end_of_turn" } }, - "metrics": null + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 67 + }, + { + "metric": "completion_tokens", + "unit": null, + "value": 37 + }, + { + "metric": "total_tokens", + "unit": null, + "value": 104 + } + ] } } ], @@ -26180,32 +30482,7 @@ "__module__": "llama_stack.apis.common.content_types", "value": "in_progress" }, - "tool_call": "\": {\"query\": \"Perplexity", - "type": "tool_call" - }, - "event_type": { - "__enum__": "ChatCompletionResponseEventType", - "__module__": "llama_stack.apis.inference.inference", - "value": "progress" - }, - "logprobs": null, - "stop_reason": null - }, - "metrics": null - } - }, - { - "__module__": "llama_stack.apis.inference.inference", - "__pydantic__": "ChatCompletionResponseStreamChunk", - "data": { - "event": { - "delta": { - "parse_status": { - "__enum__": "ToolCallParseStatus", - "__module__": "llama_stack.apis.common.content_types", - "value": "in_progress" - }, - "tool_call": " company founding date\"}}", + "tool_call": "\": {\"query\": \"Perplexity company founding date\"}}", "type": "tool_call" }, "event_type": { @@ -26234,7 +30511,7 @@ "arguments": { "query": "Perplexity company founding date" }, - "call_id": "94a9fd55-7658-482d-8595-d2c2a23b3a1e", + "call_id": "06c95bef-9b2d-4380-bf16-e1338bb7cf2c", "tool_name": "knowledge_search" }, "type": "tool_call" @@ -26275,7 +30552,23 @@ "value": "end_of_turn" } }, - "metrics": null + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 29 + }, + { + "metric": "completion_tokens", + "unit": null, + "value": 10 + }, + { + "metric": "total_tokens", + "unit": null, + "value": 39 + } + ] } } ], @@ -26438,7 +30731,7 @@ "data": { "event": { "delta": { - "text": "{\"", + "text": "The", "type": "text" }, "event_type": { @@ -26458,7 +30751,7 @@ "data": { "event": { "delta": { - "text": "type\": \"function\", \"name\": \"knowledge_search\", \"parameters", + "text": " NBA was created on August 3, 1949, with", "type": "text" }, "event_type": { @@ -26478,7 +30771,7 @@ "data": { "event": { "delta": { - "text": "\": {\"query\": \"when was the nba created\"}}", + "text": " the merger of the Basketball Association of America (BAA) and", "type": "text" }, "event_type": { @@ -26498,19 +30791,8 @@ "data": { "event": { "delta": { - "parse_status": { - "__enum__": "ToolCallParseStatus", - "__module__": "llama_stack.apis.common.content_types", - "value": "succeeded" - }, - "tool_call": { - "arguments": { - "query": "when was the nba created" - }, - "call_id": "7b01a40d-a6a8-4c86-b91d-1790e7480e57", - "tool_name": "knowledge_search" - }, - "type": "tool_call" + "text": " the National Basketball League (NBL).", + "type": "text" }, "event_type": { "__enum__": "ChatCompletionResponseEventType", @@ -26518,11 +30800,7 @@ "value": "progress" }, "logprobs": null, - "stop_reason": { - "__enum__": "StopReason", - "__module__": "llama_stack.models.llama.datatypes", - "value": "end_of_turn" - } + "stop_reason": null }, "metrics": null } @@ -26548,7 +30826,23 @@ "value": "end_of_turn" } }, - "metrics": null + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 65 + }, + { + "metric": "completion_tokens", + "unit": null, + "value": 45 + }, + { + "metric": "total_tokens", + "unit": null, + "value": 110 + } + ] } } ], @@ -26612,7 +30906,7 @@ "__module__": "llama_stack.apis.common.content_types", "value": "in_progress" }, - "tool_call": "{\"type\": \"function\", \"name\": \"knowledge_search\",", + "tool_call": "{\"type\": \"function\", \"", "type": "tool_call" }, "event_type": { @@ -26637,7 +30931,7 @@ "__module__": "llama_stack.apis.common.content_types", "value": "in_progress" }, - "tool_call": " \"parameters\": {\"query\": \"when was the nba created", + "tool_call": "name\": \"knowledge_search\", \"parameters\": {\"query\": \"", "type": "tool_call" }, "event_type": { @@ -26662,7 +30956,7 @@ "__module__": "llama_stack.apis.common.content_types", "value": "in_progress" }, - "tool_call": "\"}}", + "tool_call": "when was the nba created\"}}", "type": "tool_call" }, "event_type": { @@ -26691,7 +30985,7 @@ "arguments": { "query": "when was the nba created" }, - "call_id": "bbaf750a-0337-4c83-9bf2-76c2f72d45c3", + "call_id": "b08bb4c0-c0a1-4063-b110-3947559e4061", "tool_name": "knowledge_search" }, "type": "tool_call" @@ -26732,7 +31026,23 @@ "value": "end_of_turn" } }, - "metrics": null + "metrics": [ + { + "metric": "prompt_tokens", + "unit": null, + "value": 27 + }, + { + "metric": "completion_tokens", + "unit": null, + "value": 10 + }, + { + "metric": "total_tokens", + "unit": null, + "value": 37 + } + ] } } ], diff --git a/tests/integration/fixtures/recorded_responses/invoke_tool.json b/tests/integration/fixtures/recorded_responses/invoke_tool.json index 76191e992..30a132904 100644 --- a/tests/integration/fixtures/recorded_responses/invoke_tool.json +++ b/tests/integration/fixtures/recorded_responses/invoke_tool.json @@ -5,7 +5,7 @@ "__module__": "llama_stack.apis.tools.tools", "__pydantic__": "ToolInvocationResult", "data": { - "content": "completed\n[stderr]\nTraceback (most recent call last):\n line 5, in \n from bwrap.core import main\nModuleNotFoundError: No module named 'bwrap.core'\n[/stderr]", + "content": "completed\n[stdout]\n541\n[/stdout]", "error_code": null, "error_message": null, "metadata": null @@ -31,7 +31,7 @@ "__module__": "llama_stack.apis.tools.tools", "__pydantic__": "ToolInvocationResult", "data": { - "content": "completed\n[stderr]\nTraceback (most recent call last):\n line 5, in \n from bwrap.core import main\nModuleNotFoundError: No module named 'bwrap.core'\n[/stderr]", + "content": "completed\n[stdout]\nNumber of rows and columns in the data: (10, 13)\nColumns of the data are: 13\nColumns of the data are: Index(['Year', 'Jan', 'Feb', 'Mar', 'Apr', 'May', 'Jun', 'Jul', 'Aug', 'Sep',\n 'Oct', 'Nov', 'Dec'],\n dtype='object')\nDatatype of the columns are: Year int64\nJan float64\nFeb float64\nMar float64\nApr float64\nMay float64\nJun float64\nJul float64\nAug float64\nSep float64\nOct float64\nNov float64\nDec float64\ndtype: object\n[/stdout]", "error_code": null, "error_message": null, "metadata": null @@ -70,7 +70,7 @@ "__module__": "llama_stack.apis.tools.tools", "__pydantic__": "ToolInvocationResult", "data": { - "content": "completed\n[stderr]\nTraceback (most recent call last):\n line 5, in \n from bwrap.core import main\nModuleNotFoundError: No module named 'bwrap.core'\n[/stderr]", + "content": "completed\n[stdout]\nYear Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec\n0 2014 1.6 1.6 1.7 1.8 2.0 1.9 1.9 1.7 1.7 1.8 1.7 1.6\n1 2015 1.6 1.7 1.8 1.8 1.7 1.8 1.8 1.8 1.9 1.9 2.0 2.1\n2 2016 2.2 2.3 2.2 2.1 2.2 2.2 2.2 2.3 2.2 2.1 2.1 2.2\n3 2017 2.3 2.2 2.0 1.9 1.7 1.7 1.7 1.7 1.7 1.8 1.7 1.8\n4 2018 1.8 1.8 2.1 2.1 2.2 2.3 2.4 2.2 2.2 2.1 2.2 2.2\n[/stdout]", "error_code": null, "error_message": null, "metadata": null @@ -83,7 +83,7 @@ "__module__": "llama_stack.apis.tools.tools", "__pydantic__": "ToolInvocationResult", "data": { - "content": "completed\n[stderr]\nTraceback (most recent call last):\n line 5, in \n from bwrap.core import main\nModuleNotFoundError: No module named 'bwrap.core'\n[/stderr]", + "content": "completed\n[stderr]\nTraceback (most recent call last):\n line 142, in \n line 23, in \n from .code_execution import CodeExecutionContext, CodeExecutionRequest, CodeExecutor\nImportError: attempted relative import with no known parent package\n[/stderr]", "error_code": null, "error_message": null, "metadata": null @@ -116,6 +116,19 @@ } } }, + "[[], {\"kwargs\": {\"code\": \"import pandas as pd\\nimport matplotlib.pyplot as plt\\n\\n# Load data\\ndf = pd.read_csv(\\\"\")\\n\\n# Calculate average yearly inflation\\ndf['Average'] = df[['Jan', 'Feb', 'Mar', 'Apr', 'May', 'Jun', 'Jul', 'Aug', 'Sep', 'Oct', 'Nov', 'Dec']].mean(axis=1)\\n\\n# Plot time series\\nplt.figure(figsize=(10,6))\\nplt.plot(df['Year'], df['Average'])\\nplt.xlabel('Year')\\nplt.ylabel('Average Yearly Inflation')\\nplt.title('Average Yearly Inflation Over Time')\\nplt.grid(True)\\nplt.show()\", \"session_id\": \"\"}, \"tool_name\": \"code_interpreter\"}]": { + "type": "value", + "value": { + "__module__": "llama_stack.apis.tools.tools", + "__pydantic__": "ToolInvocationResult", + "data": { + "content": "completed", + "error_code": null, + "error_message": null, + "metadata": null + } + } + }, "[[], {\"kwargs\": {\"code\": \"import pandas as pd\\nimport matplotlib.pyplot as plt\\n\\n# Load data\\ndf = pd.read_csv(\\\"inflation.csv\\\")\\n\\n# Convert date column to datetime\\ndf['date'] = pd.to_datetime(df['date'])\\n\\n# Group by year and calculate average inflation\\naverage_inflation = df.groupby(df['date'].dt.year)['inflation'].mean()\\n\\n# Plot average yearly inflation as a time series\\nplt.figure(figsize=(10,6))\\nplt.plot(average_inflation.index, average_inflation.values, marker='o')\\nplt.title('Average Yearly Inflation')\\nplt.xlabel('Year')\\nplt.ylabel('Average Inflation')\\nplt.grid(True)\\nplt.show()\", \"session_id\": \"\"}, \"tool_name\": \"code_interpreter\"}]": { "type": "value", "value": { @@ -154,23 +167,23 @@ "type": "text" }, { - "text": "Result 1:\nDocument_id:961ff\nContent: .. _lora_finetune_label:\n\n============================\nFine-Tuning Llama2 with LoRA\n============================\n\nThis guide will teach you about `LoRA `_, a parameter-efficient finetuning technique,\nand show you how you can use torchtune to finetune a Llama2 model with LoRA.\nIf you already know what LoRA is and want to get straight to running\nyour own LoRA finetune in torchtune, you can jump to :ref:`LoRA finetuning recipe in torchtune`.\n\n.. grid:: 2\n\n .. grid-item-card:: :octicon:`mortar-board;1em;` What you will learn\n\n * What LoRA is and how it saves memory during finetuning\n * An overview of LoRA components in torchtune\n * How to run a LoRA finetune using torchtune\n * How to experiment with different LoRA configurations\n\n .. grid-item-card:: :octicon:`list-unordered;1em;` Prerequisites\n\n * Be familiar with :ref:`torchtune`\n * Make sure to :ref:`install torchtune`\n * Make sure you have downloaded the :ref:`Llama2-7B model weights`\n\nWhat is LoRA?\n-------------\n\n`LoRA `_ is an adapter-based method for\nparameter-efficient finetuning that adds trainable low-rank decomposition matrices to different layers of a neural network,\nthen freezes the network's remaining parameters. LoRA is most commonly applied to\ntransformer models, in which case it is common to add the low-rank matrices\nto some of the linear projections in each transformer layer's self-attention.\n\n.. note::\n\n If you're unfamiliar, check out these references for the `definition of rank `_\n and discussion of `low-rank approximations `_.\n\nBy finetuning with LoRA (as opposed to finetuning all model parameters),\nyou can expect to see memory savings due to a substantial reduction in the\nnumber of parameters with gradients. When using an optimizer with momentum,\nlike `AdamW `_, a parameter-efficient finetuning technique,\nand show you how you can use torchtune to finetune a Llama2 model with LoRA.\nIf you already know what LoRA is and want to get straight to running\nyour own LoRA finetune in torchtune, you can jump to :ref:`LoRA finetuning recipe in torchtune`.\n\n.. grid:: 2\n\n .. grid-item-card:: :octicon:`mortar-board;1em;` What you will learn\n\n * What LoRA is and how it saves memory during finetuning\n * An overview of LoRA components in torchtune\n * How to run a LoRA finetune using torchtune\n * How to experiment with different LoRA configurations\n\n .. grid-item-card:: :octicon:`list-unordered;1em;` Prerequisites\n\n * Be familiar with :ref:`torchtune`\n * Make sure to :ref:`install torchtune`\n * Make sure you have downloaded the :ref:`Llama2-7B model weights`\n\nWhat is LoRA?\n-------------\n\n`LoRA `_ is an adapter-based method for\nparameter-efficient finetuning that adds trainable low-rank decomposition matrices to different layers of a neural network,\nthen freezes the network's remaining parameters. LoRA is most commonly applied to\ntransformer models, in which case it is common to add the low-rank matrices\nto some of the linear projections in each transformer layer's self-attention.\n\n.. note::\n\n If you're unfamiliar, check out these references for the `definition of rank `_\n and discussion of `low-rank approximations `_.\n\nBy finetuning with LoRA (as opposed to finetuning all model parameters),\nyou can expect to see memory savings due to a substantial reduction in the\nnumber of parameters with gradients. When using an optimizer with momentum,\nlike `AdamW ` alone will not handle the definition of which parameters are trainable.\n See :ref:`below` for how to do this.\n\nLet's inspect each of these models a bit more closely.\n\n.. code-block:: bash\n\n # Print the first layer's self-attention in the usual Llama2 model\n >>> print(base_model.layers[0].attn)\n MultiHeadAttention(\n (q_proj): Linear(in_features=4096, out_features=4096, bias=False)\n (k_proj): Linear(in_features=4096, out_features=4096, bias=False)\n (v_proj): Linear(in_features=4096, out_features=4096, bias=False)\n (output_proj): Linear(in_features=4096, out_features=4096, bias=False)\n (pos_embeddings): RotaryPositionalEmbeddings()\n )\n\n # Print the same for Llama2 with LoRA weights\n >>> print(lora_model.layers[0].attn)\n MultiHeadAttention(\n (q_proj): LoRALinear(\n (dropout): Dropout(p=0.0, inplace=False)\n \n", + "text": "Result 2:\nDocument_id:cc255\nContent: LoRA to Llama2 models\n------------------------------\n\nWith torchtune, we can easily apply LoRA to Llama2 with a variety of different configurations.\nLet's take a look at how to construct Llama2 models in torchtune with and without LoRA.\n\n.. code-block:: python\n\n from torchtune.models.llama2 import llama2_7b, lora_llama2_7b\n\n # Build Llama2 without any LoRA layers\n base_model = llama2_7b()\n\n # The default settings for lora_llama2_7b will match those for llama2_7b\n # We just need to define which layers we want LoRA applied to.\n # Within each self-attention, we can choose from [\"q_proj\", \"k_proj\", \"v_proj\", and \"output_proj\"].\n # We can also set apply_lora_to_mlp=True or apply_lora_to_output=True to apply LoRA to other linear\n # layers outside of the self-attention.\n lora_model = lora_llama2_7b(lora_attn_modules=[\"q_proj\", \"v_proj\"])\n\n.. note::\n\n Calling :func:`lora_llama_2_7b ` alone will not handle the definition of which parameters are trainable.\n See :ref:`below` for how to do this.\n\nLet's inspect each of these models a bit more closely.\n\n.. code-block:: bash\n\n # Print the first layer's self-attention in the usual Llama2 model\n >>> print(base_model.layers[0].attn)\n MultiHeadAttention(\n (q_proj): Linear(in_features=4096, out_features=4096, bias=False)\n (k_proj): Linear(in_features=4096, out_features=4096, bias=False)\n (v_proj): Linear(in_features=4096, out_features=4096, bias=False)\n (output_proj): Linear(in_features=4096, out_features=4096, bias=False)\n (pos_embeddings): RotaryPositionalEmbeddings()\n )\n\n # Print the same for Llama2 with LoRA weights\n >>> print(lora_model.layers[0].attn)\n MultiHeadAttention(\n (q_proj): LoRALinear(\n (dropout): Dropout(p=0.0, inplace=False)\n \n", "type": "text" }, { - "text": "Result 3:\nDocument_id:961ff\nContent: 06% of all params are trainable.\n\n.. note::\n If you are directly using the LoRA recipe (as detailed :ref:`here`), you need only pass the\n relevant checkpoint path. Loading model weights and setting trainable parameters will be taken care\n of in the recipe.\n\n\n.. _lora_recipe_label:\n\nLoRA finetuning recipe in torchtune\n-----------------------------------\n\nFinally, we can put it all together and finetune a model using torchtune's `LoRA recipe `_.\nMake sure that you have first downloaded the Llama2 weights and tokenizer by following :ref:`these instructions`.\nYou can then run the following command to perform a LoRA finetune of Llama2-7B with two GPUs (each having VRAM of at least 16GB):\n\n.. code-block:: bash\n\n tune run --nnodes 1 --nproc_per_node 2 lora_finetune_distributed --config llama2/7B_lora\n\n.. note::\n Make sure to point to the location of your Llama2 weights and tokenizer. This can be done\n either by adding :code:`checkpointer.checkpoint_files=[my_model_checkpoint_path] tokenizer_checkpoint=my_tokenizer_checkpoint_path`\n or by directly modifying the :code:`7B_lora.yaml` file. See our \"\":ref:`config_tutorial_label`\" recipe\n for more details on how you can easily clone and modify torchtune configs.\n\n.. note::\n You can modify the value of :code:`nproc_per_node` depending on (a) the number of GPUs you have available,\n and (b) the memory constraints of your hardware.\n\nThe preceding command will run a LoRA finetune with torchtune's factory settings, but we may want to experiment a bit.\nLet's take a closer look at some of the :code:`lora_finetune_distributed` config.\n\n.. code-block:: yaml\n\n # Model Arguments\n model:\n _component_: lora_llama2_7b\n lora_attn_modules: ['q_proj', 'v_proj']\n lora_rank: 8\n lora_alpha: 16\n ...\n\nWe see that the\n", + "text": "Result 3:\nDocument_id:cc255\nContent: 06% of all params are trainable.\n\n.. note::\n If you are directly using the LoRA recipe (as detailed :ref:`here`), you need only pass the\n relevant checkpoint path. Loading model weights and setting trainable parameters will be taken care\n of in the recipe.\n\n\n.. _lora_recipe_label:\n\nLoRA finetuning recipe in torchtune\n-----------------------------------\n\nFinally, we can put it all together and finetune a model using torchtune's `LoRA recipe `_.\nMake sure that you have first downloaded the Llama2 weights and tokenizer by following :ref:`these instructions`.\nYou can then run the following command to perform a LoRA finetune of Llama2-7B with two GPUs (each having VRAM of at least 16GB):\n\n.. code-block:: bash\n\n tune run --nnodes 1 --nproc_per_node 2 lora_finetune_distributed --config llama2/7B_lora\n\n.. note::\n Make sure to point to the location of your Llama2 weights and tokenizer. This can be done\n either by adding :code:`checkpointer.checkpoint_files=[my_model_checkpoint_path] tokenizer_checkpoint=my_tokenizer_checkpoint_path`\n or by directly modifying the :code:`7B_lora.yaml` file. See our \"\":ref:`config_tutorial_label`\" recipe\n for more details on how you can easily clone and modify torchtune configs.\n\n.. note::\n You can modify the value of :code:`nproc_per_node` depending on (a) the number of GPUs you have available,\n and (b) the memory constraints of your hardware.\n\nThe preceding command will run a LoRA finetune with torchtune's factory settings, but we may want to experiment a bit.\nLet's take a closer look at some of the :code:`lora_finetune_distributed` config.\n\n.. code-block:: yaml\n\n # Model Arguments\n model:\n _component_: lora_llama2_7b\n lora_attn_modules: ['q_proj', 'v_proj']\n lora_rank: 8\n lora_alpha: 16\n ...\n\nWe see that the\n", "type": "text" }, { - "text": "Result 4:\nDocument_id:961ff\nContent: from our Llama2\nmodel without any wrappers or custom checkpoint conversion logic.\n\n.. code-block:: python\n\n # Assuming that base_model already has the pretrained Llama2 weights,\n # this will directly load them into your LoRA model without any conversion necessary.\n lora_model.load_state_dict(base_model.state_dict(), strict=False)\n\n.. note::\n Whenever loading weights with :code:`strict=False`, you should verify that any missing or extra keys in\n the loaded :code:`state_dict` are as expected. torchtune's LoRA recipes do this by default via\n :func:`validate_missing_and_unexpected_for_lora() `.\n\nOnce we've loaded the base model weights, we also want to set only LoRA parameters to trainable.\n\n.. _setting_trainable_params:\n\n.. code-block:: python\n\n from torchtune.modules.peft.peft_utils import get_adapter_params, set_trainable_params\n\n # Fetch all params from the model that are associated with LoRA.\n lora_params = get_adapter_params(lora_model)\n\n # Set requires_grad=True on lora_params, and requires_grad=False on all others.\n set_trainable_params(lora_model, lora_params)\n\n # Print the total number of parameters\n total_params = sum([p.numel() for p in lora_model.parameters()])\n trainable_params = sum([p.numel() for p in lora_model.parameters() if p.requires_grad])\n print(\n f\"\"\"\n {total_params} total params,\n {trainable_params}\" trainable params,\n {(100.0 * trainable_params / total_params):.2f}% of all params are trainable.\n \"\"\"\n )\n\n 6742609920 total params,\n 4194304 trainable params,\n 0.06% of all params are trainable.\n\n.. note::\n If you are directly using the LoRA recipe (as detailed :ref:`here`), you need only pass the\n relevant checkpoint path. Loading model weights and setting trainable parameters will be taken care\n of in the recipe.\n\n\n.. _lora_recipe_label:\n\nLoRA finetuning recipe in torchtune\n-----------------------------------\n\nFinally, we can put it all together and finetune a model using torchtune's `LoRA recipe `.\n\nOnce we've loaded the base model weights, we also want to set only LoRA parameters to trainable.\n\n.. _setting_trainable_params:\n\n.. code-block:: python\n\n from torchtune.modules.peft.peft_utils import get_adapter_params, set_trainable_params\n\n # Fetch all params from the model that are associated with LoRA.\n lora_params = get_adapter_params(lora_model)\n\n # Set requires_grad=True on lora_params, and requires_grad=False on all others.\n set_trainable_params(lora_model, lora_params)\n\n # Print the total number of parameters\n total_params = sum([p.numel() for p in lora_model.parameters()])\n trainable_params = sum([p.numel() for p in lora_model.parameters() if p.requires_grad])\n print(\n f\"\"\"\n {total_params} total params,\n {trainable_params}\" trainable params,\n {(100.0 * trainable_params / total_params):.2f}% of all params are trainable.\n \"\"\"\n )\n\n 6742609920 total params,\n 4194304 trainable params,\n 0.06% of all params are trainable.\n\n.. note::\n If you are directly using the LoRA recipe (as detailed :ref:`here`), you need only pass the\n relevant checkpoint path. Loading model weights and setting trainable parameters will be taken care\n of in the recipe.\n\n\n.. _lora_recipe_label:\n\nLoRA finetuning recipe in torchtune\n-----------------------------------\n\nFinally, we can put it all together and finetune a model using torchtune's `LoRA recipe \"}, \"tool_name\": \"web_search\"}]": { + "type": "value", + "value": { + "__module__": "llama_stack.apis.tools.tools", + "__pydantic__": "ToolInvocationResult", + "data": { + "content": "{\"query\": \"Meta founder\", \"top_k\": [{\"title\": \"Mark Zuckerberg, Founder, Chairman and Chief Executive Officer - Meta\", \"url\": \"https://about.meta.com/media-gallery/executives/mark-zuckerberg/\", \"content\": \"Mark Zuckerberg, Founder, Chairman and Chief Executive Officer | Meta Meta Quest Ray-Ban Meta Meta Horizon Meta AI Meta Verified Meta Pay Meta Horizon Workrooms Meta and you Learn about our community Shop Meta Meta Quest Meta Portal Meta Horizon Mark Zuckerberg is the founder, chairman and CEO of Meta, which he originally founded as Facebook in 2004. In October 2021, Facebook rebranded to Meta to reflect all of its products and services across its family of apps and a focus on developing social experiences for the metaverse \\u2014 moving beyond 2D screens toward immersive experiences like augmented and virtual reality to help build the next evolution in social technology. Shop Ray-Ban Meta glassesRay-Ban StoriesPrivacy informationSupported countries \\u00a9 2025 Meta\", \"score\": 0.81595254, \"raw_content\": null}, {\"title\": \"Executives - Meta\", \"url\": \"https://about.meta.com/media-gallery/executives/\", \"content\": \"Mark Zuckerberg, Founder, Chairman and Chief Executive Officer Joel Kaplan, Chief Global Affairs Officer Susan Li, Chief Financial Officer Javier Olivan, Chief Operating Officer Chris Cox, Chief Product Officer Andrew \\u2018Boz\\u2019 Bosworth, Chief Technology Officer Jennifer Newstead, Chief Legal Officer Dave Wehner, Chief Strategy Officer Will Cathcart, Head of WhatsApp Naomi Gleit, Head of Product John Hegeman, Chief Revenue Officer Adam Mosseri, Head of Instagram Erin Egan, Chief Privacy Officer, Policy Michel Protti, Chief Privacy Officer, Product Alex Schultz, Chief Marketing Officer and VP of Analytics Tom Alison, Head of Facebook Nicola Mendelsohn, Head of Global Business Group Ahmad Al-Dahle, VP and Head of GenAI at Meta Joelle Pineau, Vice President of AI Research and Head of FAIR at Meta\", \"score\": 0.70726365, \"raw_content\": null}, {\"title\": \"Meta - Leadership & Governance\", \"url\": \"https://investor.atmeta.com/leadership-and-governance/\", \"content\": \"Mr. Andreessen was a co-founder of Netscape Communications Corporation, a software company, serving in various positions, including Chief Technology Officer and Executive Vice President of Products. Ms. Killefer also served as Assistant Secretary for Management, Chief Financial Officer, and Chief Operating Officer of the U.S. Department of the Treasury from 1997 to 2000 and as a member of the IRS Oversight Board from 2000 to 2005, including as Chair of the IRS Oversight Board from 2002 to 2004. Ms. Travis has served as Executive Vice President and Chief Financial Officer of The Estee Lauder Companies Inc., a global manufacturer and marketer of skin care, makeup, fragrance and hair care products, since August 2012.\", \"score\": 0.467308, \"raw_content\": null}, {\"title\": \"Meta Platforms - Wikipedia\", \"url\": \"https://en.wikipedia.org/wiki/Meta_Platforms\", \"content\": \"Following a period of intense scrutiny and damaging whistleblower leaks, news started to emerge on October 21, 2021, about Facebook's plan to rebrand the company and change its name.[15][54] In the Q3 2021 Earnings Call on October 25, Mark Zuckerberg discussed the ongoing criticism of the company's social services and the way it operates, and pointed to the pivoting efforts to building the metaverse \\u2013 without mentioning the rebranding and the name change.[55] The metaverse vision and the name change from Facebook, Inc. to Meta Platforms was introduced at Facebook Connect on October 28, 2021.[16] Based on Facebook's PR campaign, the name change reflects the company's shifting long term focus of building the metaverse, a digital extension of the physical world by social media, virtual reality and augmented reality features.[16][56]\", \"score\": 0.14999175, \"raw_content\": null}, {\"title\": \"Mark Zuckerberg - Wikipedia\", \"url\": \"https://en.wikipedia.org/wiki/Mark_Zuckerberg\", \"content\": \"They began dating in 2003.[175] In September 2010, Chan, who was a medical student at the University of California, San Francisco at the time,[176] moved into his rented house in Palo Alto, California.[177][178] They married on May 19, 2012, in the grounds of his mansion in an event that also celebrated her graduation from medical school.[179][180] Zuckerberg revealed in July 2015 that they were expecting a baby girl and that Chan had previously experienced three miscarriages.[181] Their first daughter was born in December 2015.[182] They announced in a Chinese New Year video that their daughter's Chinese name is Chen Mingyu (Chinese: \\u9648\\u660e\\u5b87).[183] Their second daughter was born in August 2017.[184] Zuckerberg and his wife welcomed their third daughter in March 2023 and announced the news across his social media pages.[185] The couple also have a Puli dog named Beast,[186] who has over two million followers on Facebook.[187] Zuckerberg commissioned the visual artist Daniel Arsham to build a 7-foot-tall sculpture of his wife, which was unveiled in 2024.[188]\", \"score\": 0.03678684, \"raw_content\": null}]}", + "error_code": null, + "error_message": null, + "metadata": null + } + } + }, "[[], {\"kwargs\": {\"query\": \"NBA creation date\", \"session_id\": \"\", \"vector_db_ids\": [\"test-vector-db-\"]}, \"tool_name\": \"knowledge_search\"}]": { "type": "value", "value": { @@ -374,23 +400,23 @@ "type": "text" }, { - "text": "Result 1:\nDocument_id:24443\nContent: conversational data, :func:`~torchtune.datasets.chat_dataset` seems to be a good fit. For any\ncustom local dataset we always need to specify ``source``, ``data_files``, and ``split`` for any dataset\nbuilder in torchtune. For :func:`~torchtune.datasets.chat_dataset`, we additionally need to specify\n``conversation_column`` and ``conversation_style``. Our data follows the ``\"sharegpt\"`` format, so\nwe can specify that here. Altogether, our :func:`~torchtune.datasets.chat_dataset` call should\nlook like so:\n\n.. code-block:: python\n\n from torchtune.datasets import chat_dataset\n from torchtune.models.llama3 import llama3_tokenizer\n\n tokenizer = llama3_tokenizer(\"/tmp/Meta-Llama-3-8B-Instruct/original/tokenizer.model\")\n ds = chat_dataset(\n tokenizer=tokenizer,\n source=\"json\",\n data_files=\"data/my_data.json\",\n split=\"train\",\n conversation_column=\"dialogue\",\n conversation_style=\"sharegpt\",\n )\n\n.. code-block:: yaml\n\n # In config\n tokenizer:\n _component_: torchtune.models.llama3.llama3_tokenizer\n path: /tmp/Meta-Llama-3-8B-Instruct/original/tokenizer.model\n\n dataset:\n _component_: torchtune.datasets.chat_dataset\n source: json\n data_files: data/my_data.json\n split: train\n conversation_column: dialogue\n conversation_style: sharegpt\n\n.. note::\n You can pass in any keyword argument for `load_dataset `_ into all our\n Dataset classes and they will honor them. This is useful for common parameters\n such as specifying the data split with :code:`split` or configuration with\n :code:`name`\n\nIf you needed to add a prompt template, you would simply pass it into the tokenizer.\nSince we're fine-tuning Llama3, the tokenizer will handle all formatting for\nus and prompt templates are optional. Other models such as Mistral's :class:`~torchtune.models.mistral._tokenizer.MistralTokenizer`,\nuse a chat template by default (:class:`~torchtune.models.mistral.MistralChatTemplate`) to format\nall messages according to their `recommendations `_ into all our\n Dataset classes and they will honor them. This is useful for common parameters\n such as specifying the data split with :code:`split` or configuration with\n :code:`name`\n\nIf you needed to add a prompt template, you would simply pass it into the tokenizer.\nSince we're fine-tuning Llama3, the tokenizer will handle all formatting for\nus and prompt templates are optional. Other models such as Mistral's :class:`~torchtune.models.mistral._tokenizer.MistralTokenizer`,\nuse a chat template by default (:class:`~torchtune.models.mistral.MistralChatTemplate`) to format\nall messages according to their `recommendations `_, a parameter-efficient finetuning technique,\nand show you how you can use torchtune to finetune a Llama2 model with LoRA.\nIf you already know what LoRA is and want to get straight to running\nyour own LoRA finetune in torchtune, you can jump to :ref:`LoRA finetuning recipe in torchtune`.\n\n.. grid:: 2\n\n .. grid-item-card:: :octicon:`mortar-board;1em;` What you will learn\n\n * What LoRA is and how it saves memory during finetuning\n * An overview of LoRA components in torchtune\n * How to run a LoRA finetune using torchtune\n * How to experiment with different LoRA configurations\n\n .. grid-item-card:: :octicon:`list-unordered;1em;` Prerequisites\n\n * Be familiar with :ref:`torchtune`\n * Make sure to :ref:`install torchtune`\n * Make sure you have downloaded the :ref:`Llama2-7B model weights`\n\nWhat is LoRA?\n-------------\n\n`LoRA `_ is an adapter-based method for\nparameter-efficient finetuning that adds trainable low-rank decomposition matrices to different layers of a neural network,\nthen freezes the network's remaining parameters. LoRA is most commonly applied to\ntransformer models, in which case it is common to add the low-rank matrices\nto some of the linear projections in each transformer layer's self-attention.\n\n.. note::\n\n If you're unfamiliar, check out these references for the `definition of rank `_\n and discussion of `low-rank approximations `_.\n\nBy finetuning with LoRA (as opposed to finetuning all model parameters),\nyou can expect to see memory savings due to a substantial reduction in the\nnumber of parameters with gradients. When using an optimizer with momentum,\nlike `AdamW `_, a parameter-efficient finetuning technique,\nand show you how you can use torchtune to finetune a Llama2 model with LoRA.\nIf you already know what LoRA is and want to get straight to running\nyour own LoRA finetune in torchtune, you can jump to :ref:`LoRA finetuning recipe in torchtune`.\n\n.. grid:: 2\n\n .. grid-item-card:: :octicon:`mortar-board;1em;` What you will learn\n\n * What LoRA is and how it saves memory during finetuning\n * An overview of LoRA components in torchtune\n * How to run a LoRA finetune using torchtune\n * How to experiment with different LoRA configurations\n\n .. grid-item-card:: :octicon:`list-unordered;1em;` Prerequisites\n\n * Be familiar with :ref:`torchtune`\n * Make sure to :ref:`install torchtune`\n * Make sure you have downloaded the :ref:`Llama2-7B model weights`\n\nWhat is LoRA?\n-------------\n\n`LoRA `_ is an adapter-based method for\nparameter-efficient finetuning that adds trainable low-rank decomposition matrices to different layers of a neural network,\nthen freezes the network's remaining parameters. LoRA is most commonly applied to\ntransformer models, in which case it is common to add the low-rank matrices\nto some of the linear projections in each transformer layer's self-attention.\n\n.. note::\n\n If you're unfamiliar, check out these references for the `definition of rank `_\n and discussion of `low-rank approximations `_.\n\nBy finetuning with LoRA (as opposed to finetuning all model parameters),\nyou can expect to see memory savings due to a substantial reduction in the\nnumber of parameters with gradients. When using an optimizer with momentum,\nlike `AdamW `.\n.. .. _glossary_fsdp2:\n\n", + "text": "Result 3:\nDocument_id:7a06a\nContent: ` module, which we swap\n out for :class:`~torchtune.modules.peft.LoRALinear` when ``use_dora=True``.\n\n.. _glossary_distrib:\n\n\n.. TODO\n\n.. Distributed\n.. -----------\n\n.. .. _glossary_fsdp:\n\n.. Fully Sharded Data Parallel (FSDP)\n.. ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n\n.. All our ``_distributed`` recipes use `FSDP `.\n.. .. _glossary_fsdp2:\n\n", "type": "text" }, { - "text": "Result 4:\nDocument_id:961ff\nContent: 06% of all params are trainable.\n\n.. note::\n If you are directly using the LoRA recipe (as detailed :ref:`here`), you need only pass the\n relevant checkpoint path. Loading model weights and setting trainable parameters will be taken care\n of in the recipe.\n\n\n.. _lora_recipe_label:\n\nLoRA finetuning recipe in torchtune\n-----------------------------------\n\nFinally, we can put it all together and finetune a model using torchtune's `LoRA recipe `_.\nMake sure that you have first downloaded the Llama2 weights and tokenizer by following :ref:`these instructions`.\nYou can then run the following command to perform a LoRA finetune of Llama2-7B with two GPUs (each having VRAM of at least 16GB):\n\n.. code-block:: bash\n\n tune run --nnodes 1 --nproc_per_node 2 lora_finetune_distributed --config llama2/7B_lora\n\n.. note::\n Make sure to point to the location of your Llama2 weights and tokenizer. This can be done\n either by adding :code:`checkpointer.checkpoint_files=[my_model_checkpoint_path] tokenizer_checkpoint=my_tokenizer_checkpoint_path`\n or by directly modifying the :code:`7B_lora.yaml` file. See our \"\":ref:`config_tutorial_label`\" recipe\n for more details on how you can easily clone and modify torchtune configs.\n\n.. note::\n You can modify the value of :code:`nproc_per_node` depending on (a) the number of GPUs you have available,\n and (b) the memory constraints of your hardware.\n\nThe preceding command will run a LoRA finetune with torchtune's factory settings, but we may want to experiment a bit.\nLet's take a closer look at some of the :code:`lora_finetune_distributed` config.\n\n.. code-block:: yaml\n\n # Model Arguments\n model:\n _component_: lora_llama2_7b\n lora_attn_modules: ['q_proj', 'v_proj']\n lora_rank: 8\n lora_alpha: 16\n ...\n\nWe see that the\n", + "text": "Result 4:\nDocument_id:cc255\nContent: 06% of all params are trainable.\n\n.. note::\n If you are directly using the LoRA recipe (as detailed :ref:`here`), you need only pass the\n relevant checkpoint path. Loading model weights and setting trainable parameters will be taken care\n of in the recipe.\n\n\n.. _lora_recipe_label:\n\nLoRA finetuning recipe in torchtune\n-----------------------------------\n\nFinally, we can put it all together and finetune a model using torchtune's `LoRA recipe `_.\nMake sure that you have first downloaded the Llama2 weights and tokenizer by following :ref:`these instructions`.\nYou can then run the following command to perform a LoRA finetune of Llama2-7B with two GPUs (each having VRAM of at least 16GB):\n\n.. code-block:: bash\n\n tune run --nnodes 1 --nproc_per_node 2 lora_finetune_distributed --config llama2/7B_lora\n\n.. note::\n Make sure to point to the location of your Llama2 weights and tokenizer. This can be done\n either by adding :code:`checkpointer.checkpoint_files=[my_model_checkpoint_path] tokenizer_checkpoint=my_tokenizer_checkpoint_path`\n or by directly modifying the :code:`7B_lora.yaml` file. See our \"\":ref:`config_tutorial_label`\" recipe\n for more details on how you can easily clone and modify torchtune configs.\n\n.. note::\n You can modify the value of :code:`nproc_per_node` depending on (a) the number of GPUs you have available,\n and (b) the memory constraints of your hardware.\n\nThe preceding command will run a LoRA finetune with torchtune's factory settings, but we may want to experiment a bit.\nLet's take a closer look at some of the :code:`lora_finetune_distributed` config.\n\n.. code-block:: yaml\n\n # Model Arguments\n model:\n _component_: lora_llama2_7b\n lora_attn_modules: ['q_proj', 'v_proj']\n lora_rank: 8\n lora_alpha: 16\n ...\n\nWe see that the\n", "type": "text" }, { - "text": "Result 5:\nDocument_id:b49f7\nContent: etune\n:func:`torchtune.models.llama3.llama3_8b` with DoRA, you would use :func:`torchtune.models.llama3.lora_llama3_8b` with ``use_dora=True``:\n\n.. code-block:: bash\n\n tune run lora_finetune_single_device --config llama3/8B_lora_single_device \\\n model.use_dora=True\n\n.. code-block:: yaml\n\n model:\n _component_: torchtune.models.lora_llama3_8b\n use_dora: True\n\nSince DoRA extends LoRA, the parameters for :ref:`customizing LoRA ` are identical. You can also quantize the base model weights like in :ref:`glossary_qlora` by using ``quantize=True`` to reap\neven more memory savings!\n\n.. code-block:: bash\n\n tune run lora_finetune_single_device --config llama3/8B_lora_single_device \\\n model.apply_lora_to_mlp=True \\\n model.lora_attn_modules=[\"q_proj\",\"k_proj\",\"v_proj\"] \\\n model.lora_rank=16 \\\n model.lora_alpha=32 \\\n model.use_dora=True \\\n model.quantize_base=True\n\n.. code-block:: yaml\n\n model:\n _component_: torchtune.models.lora_llama3_8b\n apply_lora_to_mlp: True\n lora_attn_modules: [\"q_proj\", \"k_proj\", \"v_proj\"]\n lora_rank: 16\n lora_alpha: 32\n use_dora: True\n quantize_base: True\n\n\n.. note::\n\n Under the hood, we've enabled DoRA by adding the :class:`~torchtune.modules.peft.DoRALinear` module, which we swap\n out for :class:`~torchtune.modules.peft.LoRALinear` when ``use_dora=True``.\n\n.. _glossary_distrib:\n\n\n.. TODO\n\n.. Distributed\n.. -----------\n\n.. .. _glossary_fsdp:\n\n.. Fully Sharded Data Parallel (FSDP)\n.. ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n\n.. All our ``_distributed`` recipes use `FSDP `.\n.. .. _glossary_fsdp2:\n\n", + "text": "Result 5:\nDocument_id:7a06a\nContent: etune\n:func:`torchtune.models.llama3.llama3_8b` with DoRA, you would use :func:`torchtune.models.llama3.lora_llama3_8b` with ``use_dora=True``:\n\n.. code-block:: bash\n\n tune run lora_finetune_single_device --config llama3/8B_lora_single_device \\\n model.use_dora=True\n\n.. code-block:: yaml\n\n model:\n _component_: torchtune.models.lora_llama3_8b\n use_dora: True\n\nSince DoRA extends LoRA, the parameters for :ref:`customizing LoRA ` are identical. You can also quantize the base model weights like in :ref:`glossary_qlora` by using ``quantize=True`` to reap\neven more memory savings!\n\n.. code-block:: bash\n\n tune run lora_finetune_single_device --config llama3/8B_lora_single_device \\\n model.apply_lora_to_mlp=True \\\n model.lora_attn_modules=[\"q_proj\",\"k_proj\",\"v_proj\"] \\\n model.lora_rank=16 \\\n model.lora_alpha=32 \\\n model.use_dora=True \\\n model.quantize_base=True\n\n.. code-block:: yaml\n\n model:\n _component_: torchtune.models.lora_llama3_8b\n apply_lora_to_mlp: True\n lora_attn_modules: [\"q_proj\", \"k_proj\", \"v_proj\"]\n lora_rank: 16\n lora_alpha: 32\n use_dora: True\n quantize_base: True\n\n\n.. note::\n\n Under the hood, we've enabled DoRA by adding the :class:`~torchtune.modules.peft.DoRALinear` module, which we swap\n out for :class:`~torchtune.modules.peft.LoRALinear` when ``use_dora=True``.\n\n.. _glossary_distrib:\n\n\n.. TODO\n\n.. Distributed\n.. -----------\n\n.. .. _glossary_fsdp:\n\n.. Fully Sharded Data Parallel (FSDP)\n.. ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n\n.. All our ``_distributed`` recipes use `FSDP `.\n.. .. _glossary_fsdp2:\n\n", "type": "text" }, { @@ -402,11 +428,11 @@ "error_message": null, "metadata": { "document_ids": [ - "24443dfb-a0b3-4ce8-820e-3fb1f12364bb", - "961ff2d1-8887-41ef-a4fe-fa4cbab7b932", - "b49f7985-6615-4dcf-99be-d1765b6a6fc6", - "961ff2d1-8887-41ef-a4fe-fa4cbab7b932", - "b49f7985-6615-4dcf-99be-d1765b6a6fc6" + "16a6ae01-049e-4a44-b305-8248d20a8f7d", + "cc2559a9-2b56-43d8-9ec4-b2181bb96acb", + "7a06a3a9-7e9d-4693-8c07-15343f0654aa", + "cc2559a9-2b56-43d8-9ec4-b2181bb96acb", + "7a06a3a9-7e9d-4693-8c07-15343f0654aa" ] } }