Merge remote-tracking branch 'upstream/main' into peterj/autogenupdate0424

* upstream/main:
  Remove `name` field from OpenAI Assistant Message (microsoft#6388)
  Introduce workbench (microsoft#6340)
  TEST/change gpt4, gpt4o serise to gpt4.1nano (microsoft#6375)
  update website version (microsoft#6364)
  Add self-debugging loop to `CodeExecutionAgent` (microsoft#6306)
  Fix: deserialize model_context in AssistantAgent and SocietyOfMindAgent and CodeExecutorAgent (microsoft#6337)
  Add azure ai agent (microsoft#6191)
  Avoid re-registering a message type already registered (microsoft#6354)
  Added support for exposing GPUs to docker code executor (microsoft#6339)
  fix: ollama fails when tools use optional args (microsoft#6343)
  Add an example using autogen-core and FastAPI to create streaming responses (microsoft#6335)
  FEAT: SelectorGroupChat could using stream inner select_prompt (microsoft#6286)
  Add experimental notice to canvas (microsoft#6349)
  DOC: add extentions - autogen-oaiapi and autogen-contextplus (microsoft#6338)
  fix: ensure serialized messages are passed to LLMStreamStartEvent (microsoft#6344)
  Generalize Continuous SystemMessage merging via model_info[“multiple_system_messages”] instead of `startswith("gemini-")` (microsoft#6345)
  Agentchat canvas (microsoft#6215)

Signed-off-by: Peter Jausovec <[email protected]>

Author: peterj

.github/ISSUE_TEMPLATE/1-bug_report.yml

@@ -90,6 +90,7 @@ body:
       multiple: false
       options:
         - "Python dev (main branch)"
+        - "Python 0.5.4"
         - "Python 0.5.3"
         - "Python 0.5.2"
         - "Python 0.5.1"

.github/workflows/docs.yml

@@ -33,7 +33,7 @@ jobs:
           [
             # For main use the workflow target
             { ref: "${{github.ref}}", dest-dir: dev, uv-version: "0.5.13", sphinx-release-override: "dev" },
-            { ref: "python-v0.5.3", dest-dir: stable, uv-version: "0.5.13", sphinx-release-override: "stable" },
+            { ref: "python-v0.5.4", dest-dir: stable, uv-version: "0.5.13", sphinx-release-override: "stable" },
             { ref: "v0.4.0.post1", dest-dir: "0.4.0", uv-version: "0.5.13", sphinx-release-override: "" },
             { ref: "v0.4.1", dest-dir: "0.4.1", uv-version: "0.5.13", sphinx-release-override: "" },
             { ref: "v0.4.2", dest-dir: "0.4.2", uv-version: "0.5.13", sphinx-release-override: "" },
@@ -47,6 +47,7 @@ jobs:
             { ref: "python-v0.5.1", dest-dir: "0.5.1", uv-version: "0.5.13", sphinx-release-override: "" },
             { ref: "python-v0.5.2", dest-dir: "0.5.2", uv-version: "0.5.13", sphinx-release-override: "" },
             { ref: "python-v0.5.3", dest-dir: "0.5.3", uv-version: "0.5.13", sphinx-release-override: "" },
+            { ref: "python-v0.5.4", dest-dir: "0.5.4", uv-version: "0.5.13", sphinx-release-override: "" },
           ]
     steps:
       - name: Checkout

docs/switcher.json

@@ -5,11 +5,16 @@
         "url": "/autogen/dev/"
     },
     {
-        "name": "0.5.3 (stable)",
+        "name": "0.5.4 (stable)",
         "version": "stable",
         "url": "/autogen/stable/",
         "preferred": true
     },
+    {
+        "name": "0.5.3",
+        "version": "0.5.3",
+        "url": "/autogen/0.5.3/"
+    },
     {
         "name": "0.5.2",
         "version": "0.5.2",

python/packages/autogen-agentchat/src/autogen_agentchat/agents/_assistant_agent.py

@@ -1337,7 +1337,7 @@ def _from_config(cls, config: AssistantAgentConfig) -> Self:
             model_client=ChatCompletionClient.load_component(config.model_client),
             tools=[BaseTool.load_component(tool) for tool in config.tools] if config.tools else None,
             handoffs=config.handoffs,
-            model_context=None,
+            model_context=ChatCompletionContext.load_component(config.model_context) if config.model_context else None,
             memory=[Memory.load_component(memory) for memory in config.memory] if config.memory else None,
             description=config.description,
             system_message=config.system_message,

python/packages/autogen-agentchat/src/autogen_agentchat/agents/_code_executor_agent.py

@@ -286,6 +286,7 @@ def _to_config(self) -> SocietyOfMindAgentConfig:
             description=self.description,
             instruction=self._instruction,
             response_prompt=self._response_prompt,
+            model_context=self._model_context.dump_component(),
         )
 
     @classmethod
@@ -299,4 +300,5 @@ def _from_config(cls, config: SocietyOfMindAgentConfig) -> Self:
             description=config.description or cls.DEFAULT_DESCRIPTION,
             instruction=config.instruction or cls.DEFAULT_INSTRUCTION,
             response_prompt=config.response_prompt or cls.DEFAULT_RESPONSE_PROMPT,
+            model_context=ChatCompletionContext.load_component(config.model_context) if config.model_context else None,
         )

python/packages/autogen-agentchat/src/autogen_agentchat/agents/_society_of_mind_agent.py

@@ -433,22 +433,33 @@ def to_text(self) -> str:
 
 
 class CodeGenerationEvent(BaseAgentEvent):
-    """An event signaling code generation for execution."""
+    """An event signaling code generation event."""
+
+    retry_attempt: int
+    "Retry number, 0 means first generation"
 
     content: str
     "The complete content as string."
 
-    type: Literal["CodeGenerationEvent"] = "CodeGenerationEvent"
-
     code_blocks: List[CodeBlock]
+    "List of code blocks present in content"
+
+    type: Literal["CodeGenerationEvent"] = "CodeGenerationEvent"
 
     def to_text(self) -> str:
         return self.content
 
 
 class CodeExecutionEvent(BaseAgentEvent):
-    type: Literal["CodeExecutionEvent"] = "CodeExecutionEvent"
+    """An event signaling code execution event."""
+
+    retry_attempt: int
+    "Retry number, 0 means first execution"
+
     result: CodeResult
+    "Code Execution Result"
+
+    type: Literal["CodeExecutionEvent"] = "CodeExecutionEvent"
 
     def to_text(self) -> str:
         return self.result.output
@@ -531,6 +542,18 @@ def to_text(self) -> str:
         return str(self.content)
 
 
+class SelectorEvent(BaseAgentEvent):
+    """An event emitted from the `SelectorGroupChat`."""
+
+    content: str
+    """The content of the event."""
+
+    type: Literal["SelectorEvent"] = "SelectorEvent"
+
+    def to_text(self) -> str:
+        return str(self.content)
+
+
 class MessageFactory:
     """:meta private:
 

python/packages/autogen-agentchat/src/autogen_agentchat/messages.py

@@ -73,7 +73,8 @@ def __init__(
         for agent in participants:
             for message_type in agent.produced_message_types:
                 try:
-                    if issubclass(message_type, StructuredMessage):
+                    is_registered = self._message_factory.is_registered(message_type)  # type: ignore[reportUnknownArgumentType]
+                    if issubclass(message_type, StructuredMessage) and not is_registered:
                         self._message_factory.register(message_type)  # type: ignore[reportUnknownArgumentType]
                 except TypeError:
                     # Not a class or not a valid subclassable type (skip)

python/packages/autogen-agentchat/src/autogen_agentchat/teams/_group_chat/_base_group_chat.py

@@ -5,7 +5,14 @@
 from typing import Any, Awaitable, Callable, Dict, List, Mapping, Optional, Sequence, Union, cast
 
 from autogen_core import AgentRuntime, Component, ComponentModel
-from autogen_core.models import AssistantMessage, ChatCompletionClient, ModelFamily, SystemMessage, UserMessage
+from autogen_core.models import (
+    AssistantMessage,
+    ChatCompletionClient,
+    CreateResult,
+    ModelFamily,
+    SystemMessage,
+    UserMessage,
+)
 from pydantic import BaseModel
 from typing_extensions import Self
 
@@ -16,6 +23,8 @@
     BaseAgentEvent,
     BaseChatMessage,
     MessageFactory,
+    ModelClientStreamingChunkEvent,
+    SelectorEvent,
 )
 from ...state import SelectorManagerState
 from ._base_group_chat import BaseGroupChat
@@ -56,6 +65,7 @@ def __init__(
         max_selector_attempts: int,
         candidate_func: Optional[CandidateFuncType],
         emit_team_events: bool,
+        model_client_streaming: bool = False,
     ) -> None:
         super().__init__(
             name,
@@ -79,6 +89,7 @@ def __init__(
         self._max_selector_attempts = max_selector_attempts
         self._candidate_func = candidate_func
         self._is_candidate_func_async = iscoroutinefunction(self._candidate_func)
+        self._model_client_streaming = model_client_streaming
 
     async def validate_group_state(self, messages: List[BaseChatMessage] | None) -> None:
         pass
@@ -194,7 +205,26 @@ async def _select_speaker(self, roles: str, participants: List[str], history: st
         num_attempts = 0
         while num_attempts < max_attempts:
             num_attempts += 1
-            response = await self._model_client.create(messages=select_speaker_messages)
+            if self._model_client_streaming:
+                chunk: CreateResult | str = ""
+                async for _chunk in self._model_client.create_stream(messages=select_speaker_messages):
+                    chunk = _chunk
+                    if self._emit_team_events:
+                        if isinstance(chunk, str):
+                            await self._output_message_queue.put(
+                                ModelClientStreamingChunkEvent(content=cast(str, _chunk), source=self._name)
+                            )
+                        else:
+                            assert isinstance(chunk, CreateResult)
+                            assert isinstance(chunk.content, str)
+                            await self._output_message_queue.put(
+                                SelectorEvent(content=chunk.content, source=self._name)
+                            )
+                # The last chunk must be CreateResult.
+                assert isinstance(chunk, CreateResult)
+                response = chunk
+            else:
+                response = await self._model_client.create(messages=select_speaker_messages)
             assert isinstance(response.content, str)
             select_speaker_messages.append(AssistantMessage(content=response.content, source="selector"))
             # NOTE: we use all participant names to check for mentions, even if the previous speaker is not allowed.
@@ -281,6 +311,7 @@ class SelectorGroupChatConfig(BaseModel):
     # selector_func: ComponentModel | None
     max_selector_attempts: int = 3
     emit_team_events: bool = False
+    model_client_streaming: bool = False
 
 
 class SelectorGroupChat(BaseGroupChat, Component[SelectorGroupChatConfig]):
@@ -311,6 +342,7 @@ class SelectorGroupChat(BaseGroupChat, Component[SelectorGroupChatConfig]):
             selection using model. If the function returns an empty list or `None`, `SelectorGroupChat` will raise a `ValueError`.
             This function is only used if `selector_func` is not set. The `allow_repeated_speaker` will be ignored if set.
         emit_team_events (bool, optional): Whether to emit team events through :meth:`BaseGroupChat.run_stream`. Defaults to False.
+        model_client_streaming (bool, optional): Whether to use streaming for the model client. (This is useful for reasoning models like QwQ). Defaults to False.
 
     Raises:
         ValueError: If the number of participants is less than two or if the selector prompt is invalid.
@@ -453,6 +485,7 @@ def __init__(
         candidate_func: Optional[CandidateFuncType] = None,
         custom_message_types: List[type[BaseAgentEvent | BaseChatMessage]] | None = None,
         emit_team_events: bool = False,
+        model_client_streaming: bool = False,
     ):
         super().__init__(
             participants,
@@ -473,6 +506,7 @@ def __init__(
         self._selector_func = selector_func
         self._max_selector_attempts = max_selector_attempts
         self._candidate_func = candidate_func
+        self._model_client_streaming = model_client_streaming
 
     def _create_group_chat_manager_factory(
         self,
@@ -505,6 +539,7 @@ def _create_group_chat_manager_factory(
             self._max_selector_attempts,
             self._candidate_func,
             self._emit_team_events,
+            self._model_client_streaming,
         )
 
     def _to_config(self) -> SelectorGroupChatConfig:
@@ -518,6 +553,7 @@ def _to_config(self) -> SelectorGroupChatConfig:
             max_selector_attempts=self._max_selector_attempts,
             # selector_func=self._selector_func.dump_component() if self._selector_func else None,
             emit_team_events=self._emit_team_events,
+            model_client_streaming=self._model_client_streaming,
         )
 
     @classmethod
@@ -536,4 +572,5 @@ def _from_config(cls, config: SelectorGroupChatConfig) -> Self:
             # if config.selector_func
             # else None,
             emit_team_events=config.emit_team_events,
+            model_client_streaming=config.model_client_streaming,
         )

python/packages/autogen-agentchat/src/autogen_agentchat/teams/_group_chat/_selector_group_chat.py

@@ -0,0 +1,126 @@
+import pytest
+from autogen_agentchat.agents import (
+    AssistantAgent,
+    CodeExecutorAgent,
+    SocietyOfMindAgent,
+)
+from autogen_agentchat.teams import RoundRobinGroupChat
+from autogen_core.model_context import (
+    BufferedChatCompletionContext,
+    ChatCompletionContext,
+    HeadAndTailChatCompletionContext,
+    TokenLimitedChatCompletionContext,
+    UnboundedChatCompletionContext,
+)
+from autogen_ext.code_executors.local import LocalCommandLineCodeExecutor
+from autogen_ext.models.replay import ReplayChatCompletionClient
+
+
+@pytest.mark.parametrize(
+    "model_context_class",
+    [
+        UnboundedChatCompletionContext(),
+        BufferedChatCompletionContext(buffer_size=5),
+        TokenLimitedChatCompletionContext(model_client=ReplayChatCompletionClient([]), token_limit=5),
+        HeadAndTailChatCompletionContext(head_size=3, tail_size=2),
+    ],
+)
+def test_serialize_and_deserialize_model_context_on_assistant_agent(model_context_class: ChatCompletionContext) -> None:
+    """Test the serialization and deserialization of the message context on the AssistantAgent."""
+    agent = AssistantAgent(
+        name="assistant",
+        model_client=ReplayChatCompletionClient([]),
+        description="An assistant agent.",
+        model_context=model_context_class,
+    )
+
+    # Serialize the agent
+    serialized_agent = agent.dump_component()
+    # Deserialize the agent
+    deserialized_agent = AssistantAgent.load_component(serialized_agent)
+
+    # Check that the deserialized agent has the same model context as the original agent
+    original_model_context = agent.model_context
+    deserialized_model_context = deserialized_agent.model_context
+
+    assert isinstance(original_model_context, type(deserialized_model_context))
+    assert isinstance(deserialized_model_context, type(original_model_context))
+    assert original_model_context.dump_component() == deserialized_model_context.dump_component()
+
+
+@pytest.mark.parametrize(
+    "model_context_class",
+    [
+        UnboundedChatCompletionContext(),
+        BufferedChatCompletionContext(buffer_size=5),
+        TokenLimitedChatCompletionContext(model_client=ReplayChatCompletionClient([]), token_limit=5),
+        HeadAndTailChatCompletionContext(head_size=3, tail_size=2),
+    ],
+)
+def test_serialize_and_deserialize_model_context_on_society_of_mind_agent(
+    model_context_class: ChatCompletionContext,
+) -> None:
+    """Test the serialization and deserialization of the message context on the AssistantAgent."""
+    agent1 = AssistantAgent(
+        name="assistant1", model_client=ReplayChatCompletionClient([]), description="An assistant agent."
+    )
+    agent2 = AssistantAgent(
+        name="assistant2", model_client=ReplayChatCompletionClient([]), description="An assistant agent."
+    )
+    team = RoundRobinGroupChat(
+        participants=[agent1, agent2],
+    )
+    agent = SocietyOfMindAgent(
+        name="assistant",
+        model_client=ReplayChatCompletionClient([]),
+        description="An assistant agent.",
+        team=team,
+        model_context=model_context_class,
+    )
+
+    # Serialize the agent
+    serialized_agent = agent.dump_component()
+    # Deserialize the agent
+    deserialized_agent = SocietyOfMindAgent.load_component(serialized_agent)
+
+    # Check that the deserialized agent has the same model context as the original agent
+    original_model_context = agent.model_context
+    deserialized_model_context = deserialized_agent.model_context
+
+    assert isinstance(original_model_context, type(deserialized_model_context))
+    assert isinstance(deserialized_model_context, type(original_model_context))
+    assert original_model_context.dump_component() == deserialized_model_context.dump_component()
+
+
+@pytest.mark.parametrize(
+    "model_context_class",
+    [
+        UnboundedChatCompletionContext(),
+        BufferedChatCompletionContext(buffer_size=5),
+        TokenLimitedChatCompletionContext(model_client=ReplayChatCompletionClient([]), token_limit=5),
+        HeadAndTailChatCompletionContext(head_size=3, tail_size=2),
+    ],
+)
+def test_serialize_and_deserialize_model_context_on_code_executor_agent(
+    model_context_class: ChatCompletionContext,
+) -> None:
+    """Test the serialization and deserialization of the message context on the AssistantAgent."""
+    agent = CodeExecutorAgent(
+        name="assistant",
+        code_executor=LocalCommandLineCodeExecutor(),
+        description="An assistant agent.",
+        model_context=model_context_class,
+    )
+
+    # Serialize the agent
+    serialized_agent = agent.dump_component()
+    # Deserialize the agent
+    deserialized_agent = CodeExecutorAgent.load_component(serialized_agent)
+
+    # Check that the deserialized agent has the same model context as the original agent
+    original_model_context = agent.model_context
+    deserialized_model_context = deserialized_agent.model_context
+
+    assert isinstance(original_model_context, type(deserialized_model_context))
+    assert isinstance(deserialized_model_context, type(original_model_context))
+    assert original_model_context.dump_component() == deserialized_model_context.dump_component()