mirror of
https://github.com/meta-llama/llama-stack.git
synced 2025-06-28 02:53:30 +00:00
* Add distribution CLI scaffolding * More progress towards `llama distribution install` * getting closer to a distro definition, distro install + configure works * Distribution server now functioning * read existing configuration, save enums properly * Remove inference uvicorn server entrypoint and llama inference CLI command * updated dependency and client model name * Improved exception handling * local imports for faster cli * undo a typo, add a passthrough distribution * implement full-passthrough in the server * add safety adapters, configuration handling, server + clients * cleanup, moving stuff to common, nuke utils * Add a Path() wrapper at the earliest place * fixes * Bring agentic system api to toolchain Add adapter dependencies and resolve adapters using a topological sort * refactor to reduce size of `agentic_system` * move straggler files and fix some important existing bugs * ApiSurface -> Api * refactor a method out * Adapter -> Provider * Make each inference provider into its own subdirectory * installation fixes * Rename Distribution -> DistributionSpec, simplify RemoteProviders * dict key instead of attr * update inference config to take model and not model_dir * Fix passthrough streaming, send headers properly not part of body :facepalm * update safety to use model sku ids and not model dirs * Update cli_reference.md * minor fixes * add DistributionConfig, fix a bug in model download * Make install + start scripts do proper configuration automatically * Update CLI_reference * Nuke fp8_requirements, fold fbgemm into common requirements * Update README, add newline between API surface configurations * Refactor download functionality out of the Command so can be reused * Add `llama model download` alias for `llama download` * Show message about checksum file so users can check themselves * Simpler intro statements * get ollama working * Reduce a bunch of dependencies from toolchain Some improvements to the distribution install script * Avoid using `conda run` since it buffers everything * update dependencies and rely on LLAMA_TOOLCHAIN_DIR for dev purposes * add validation for configuration input * resort imports * make optional subclasses default to yes for configuration * Remove additional_pip_packages; move deps to providers * for inline make 8b model the default * Add scripts to MANIFEST * allow installing from test.pypi.org * Fix #2 to help with testing packages * Must install llama-models at that same version first * fix PIP_ARGS --------- Co-authored-by: Hardik Shah <hjshah@fb.com> Co-authored-by: Hardik Shah <hjshah@meta.com>
166 lines
6.3 KiB
Python
166 lines
6.3 KiB
Python
# Copyright (c) Meta Platforms, Inc. and affiliates.
|
|
# All rights reserved.
|
|
#
|
|
# This source code is licensed under the terms described in the LICENSE file in
|
|
# the root directory of this source tree.
|
|
|
|
from typing import Optional
|
|
|
|
from llama_models.llama3_1.api.datatypes import ToolResponseMessage
|
|
from llama_models.llama3_1.api.tool_utils import ToolUtils
|
|
|
|
from llama_toolchain.agentic_system.api import (
|
|
AgenticSystemTurnResponseEventType,
|
|
StepType,
|
|
)
|
|
|
|
from termcolor import cprint
|
|
|
|
|
|
class LogEvent:
|
|
def __init__(
|
|
self,
|
|
role: Optional[str] = None,
|
|
content: str = "",
|
|
end: str = "\n",
|
|
color="white",
|
|
):
|
|
self.role = role
|
|
self.content = content
|
|
self.color = color
|
|
self.end = "\n" if end is None else end
|
|
|
|
def __str__(self):
|
|
if self.role is not None:
|
|
return f"{self.role}> {self.content}"
|
|
else:
|
|
return f"{self.content}"
|
|
|
|
def print(self, flush=True):
|
|
cprint(f"{str(self)}", color=self.color, end=self.end, flush=flush)
|
|
|
|
|
|
EventType = AgenticSystemTurnResponseEventType
|
|
|
|
|
|
class EventLogger:
|
|
async def log(self, event_generator, stream=True):
|
|
previous_event_type = None
|
|
previous_step_type = None
|
|
|
|
async for chunk in event_generator:
|
|
if not hasattr(chunk, "event"):
|
|
# Need to check for custom tool first
|
|
# since it does not produce event but instead
|
|
# a Message
|
|
if isinstance(chunk, ToolResponseMessage):
|
|
yield chunk, LogEvent(
|
|
role="CustomTool", content=chunk.content, color="grey"
|
|
)
|
|
continue
|
|
|
|
event = chunk.event
|
|
event_type = event.payload.event_type
|
|
if event_type in {
|
|
EventType.turn_start.value,
|
|
EventType.turn_complete.value,
|
|
}:
|
|
# Currently not logging any turn realted info
|
|
yield event, None
|
|
continue
|
|
|
|
step_type = event.payload.step_type
|
|
# handle safety
|
|
if (
|
|
step_type == StepType.shield_call
|
|
and event_type == EventType.step_complete.value
|
|
):
|
|
response = event.payload.step_details.response
|
|
if not response.is_violation:
|
|
yield event, LogEvent(
|
|
role=step_type, content="No Violation", color="magenta"
|
|
)
|
|
else:
|
|
yield event, LogEvent(
|
|
role=step_type,
|
|
content=f"{response.violation_type} {response.violation_return_message}",
|
|
color="red",
|
|
)
|
|
|
|
# handle inference
|
|
if step_type == StepType.inference:
|
|
if stream:
|
|
if event_type == EventType.step_start.value:
|
|
# TODO: Currently this event is never received
|
|
yield event, LogEvent(
|
|
role=step_type, content="", end="", color="yellow"
|
|
)
|
|
elif event_type == EventType.step_progress.value:
|
|
# HACK: if previous was not step/event was not inference's step_progress
|
|
# this is the first time we are getting model inference response
|
|
# aka equivalent to step_start for inference. Hence,
|
|
# start with "Model>".
|
|
if (
|
|
previous_event_type != EventType.step_progress.value
|
|
and previous_step_type != StepType.inference
|
|
):
|
|
yield event, LogEvent(
|
|
role=step_type, content="", end="", color="yellow"
|
|
)
|
|
|
|
if event.payload.tool_call_delta:
|
|
if isinstance(event.payload.tool_call_delta.content, str):
|
|
yield event, LogEvent(
|
|
role=None,
|
|
content=event.payload.tool_call_delta.content,
|
|
end="",
|
|
color="cyan",
|
|
)
|
|
else:
|
|
yield event, LogEvent(
|
|
role=None,
|
|
content=event.payload.model_response_text_delta,
|
|
end="",
|
|
color="yellow",
|
|
)
|
|
else:
|
|
# step_complete
|
|
yield event, LogEvent(role=None, content="")
|
|
|
|
else:
|
|
# Not streaming
|
|
if event_type == EventType.step_complete.value:
|
|
response = event.payload.step_details.model_response
|
|
if response.tool_calls:
|
|
content = ToolUtils.encode_tool_call(response.tool_calls[0])
|
|
else:
|
|
content = response.content
|
|
yield event, LogEvent(
|
|
role=step_type,
|
|
content=content,
|
|
color="yellow",
|
|
)
|
|
|
|
# handle tool_execution
|
|
if (
|
|
step_type == StepType.tool_execution
|
|
and
|
|
# Only print tool calls and responses at the step_complete event
|
|
event_type == EventType.step_complete.value
|
|
):
|
|
details = event.payload.step_details
|
|
for t in details.tool_calls:
|
|
yield event, LogEvent(
|
|
role=step_type,
|
|
content=f"Tool:{t.tool_name} Args:{t.arguments}",
|
|
color="green",
|
|
)
|
|
for r in details.tool_responses:
|
|
yield event, LogEvent(
|
|
role=step_type,
|
|
content=f"Tool:{r.tool_name} Response:{r.content}",
|
|
color="green",
|
|
)
|
|
|
|
preivous_event_type = event_type
|
|
previous_step_type = step_type
|