revert some unintentional changes by copying source of truth to llama-models

docs/_static/llama-stack-spec.html

@@ -4163,6 +4163,11 @@
                         ]
                     },
                     "arguments": {
+                        "oneOf": [
+                            {
+                                "type": "string"
+                            },
+                            {
                                 "type": "object",
                                 "additionalProperties": {
                                     "oneOf": [
@@ -4228,6 +4233,11 @@
                                     ]
                                 }
                             }
+                        ]
+                    },
+                    "arguments_json": {
+                        "type": "string"
+                    }
                 },
                 "additionalProperties": false,
                 "required": [

docs/_static/llama-stack-spec.yaml

@@ -2890,7 +2890,9 @@ components:
               title: BuiltinTool
             - type: string
         arguments:
-          type: object
+          oneOf:
+            - type: string
+            - type: object
               additionalProperties:
                 oneOf:
                   - type: string
@@ -2914,6 +2916,8 @@ components:
                         - type: number
                         - type: boolean
                         - type: 'null'
+        arguments_json:
+          type: string
       additionalProperties: false
       required:
         - call_id

llama_stack/models/llama/datatypes.py

@@ -38,7 +38,14 @@ RecursiveType = Union[Primitive, List[Primitive], Dict[str, Primitive]]
 class ToolCall(BaseModel):
     call_id: str
     tool_name: Union[BuiltinTool, str]
-    arguments: Dict[str, RecursiveType]
+    # Plan is to deprecate the Dict in favor of a JSON string
+    # that is parsed on the client side instead of trying to manage
+    # the recursive type here.
+    # Making this a union so that client side can start prepping for this change.
+    # Eventually, we will remove both the Dict and arguments_json field,
+    # and arguments will just be a str
+    arguments: Union[str, Dict[str, RecursiveType]]
+    arguments_json: Optional[str] = None
 
     @field_validator("tool_name", mode="before")
     @classmethod

llama_stack/models/llama/llama3/quantization/__init__.py

@@ -0,0 +1,5 @@
+# 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.

llama_stack/models/llama/llama4/chat_format.py

@@ -210,9 +210,12 @@ class ChatFormat:
                 content = ToolUtils.encode_tool_call(t, tool_prompt_format)
                 _process_content(content)
 
+        # Tool calls and Tool Response messages should be eom
         eom = False
         if message.role == "assistant":
-            eom = message.stop_reason == StopReason.end_of_message
+            eom = message.stop_reason == StopReason.end_of_message or message.tool_calls
+        elif message.role == "tool":
+            eom = True
 
         tokens.append(self.tokenizer.special_tokens["<|eom|>" if eom else "<|eot|>"])
         return tokens, images
@@ -247,6 +250,11 @@ class ChatFormat:
         if content.startswith(header_str):
             content = content[len(header_str) :]
 
+        ipython = content.startswith("<|python_start|>")
+        if ipython:
+            content = content[len("<|python_start|>") :]
+            content = content.replace("<|python_end|>", "")
+
         if content.endswith("<|eot|>"):
             content = content[: -len("<|eot|>")]
             stop_reason = StopReason.end_of_turn
@@ -277,6 +285,11 @@ class ChatFormat:
                 }
                 if tool_name in BuiltinTool.__members__:
                     tool_name = BuiltinTool[tool_name]
+            elif ipython:
+                tool_name = BuiltinTool.code_interpreter
+                tool_arguments = {
+                    "code": content,
+                }
 
         tool_calls = []
         if tool_name is not None and tool_arguments is not None:

llama_stack/models/llama/llama4/generation.py

@@ -4,22 +4,6 @@
 # This source code is licensed under the terms described in the LICENSE file in
 # the root directory of this source tree.
 
-# Copyright (c) Meta Platforms, Inc. and affiliates.
-# All rights reserved.
-#
-# This source code is licensed under the terms described in the LICENSE file in
-# top-level folder for each specific model found within the models/ directory at
-# the top-level of this source tree.
-
-# 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.
-
-# Copyright (c) Meta Platforms, Inc. and affiliates.
-# This software may be used and distributed in accordance with the terms of the Llama 3 Community License Agreement.
-
 import codecs
 import io
 import json

llama_stack/models/llama/llama4/quantization/__init__.py

@@ -0,0 +1,5 @@
+# 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.

llama_stack/models/llama/llama4/tokenizer.py

@@ -4,16 +4,6 @@
 # This source code is licensed under the terms described in the LICENSE file in
 # the root directory of this source tree.
 
-# Copyright (c) Meta Platforms, Inc. and affiliates.
-# All rights reserved.
-#
-# This source code is licensed under the terms described in the LICENSE file in
-# top-level folder for each specific model found within the models/ directory at
-# the top-level of this source tree.
-
-# Copyright (c) Meta Platforms, Inc. and affiliates.
-# This software may be used and distributed in accordance with the terms of the Llama 3 Community License Agreement.
-
 import os
 from logging import getLogger
 from pathlib import Path

llama_stack/providers/remote/inference/vllm/vllm.py

@@ -96,6 +96,7 @@ def _convert_to_vllm_tool_calls_in_response(
             call_id=call.id,
             tool_name=call.function.name,
             arguments=json.loads(call.function.arguments),
+            arguments_json=call.function.arguments,
         )
         for call in tool_calls
     ]
@@ -175,6 +176,7 @@ async def _process_vllm_chat_completion_stream_response(
                                 call_id=tool_call_buf.call_id,
                                 tool_name=tool_call_buf.tool_name,
                                 arguments=args,
+                                arguments_json=args_str,
                             ),
                             parse_status=ToolCallParseStatus.succeeded,
                         ),