From e0375962284525aeb077b8553bdae1f966babc83 Mon Sep 17 00:00:00 2001 From: Griffin Bassman Date: Thu, 21 Nov 2024 19:17:30 -0500 Subject: [PATCH 01/10] typo: agbench readme (#4302) --- python/packages/agbench/README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/python/packages/agbench/README.md b/python/packages/agbench/README.md index e0b9c1c84694..a8209a1e9d25 100644 --- a/python/packages/agbench/README.md +++ b/python/packages/agbench/README.md @@ -10,7 +10,7 @@ If you are already an AutoGenBench pro, and want the full technical specificatio ## Docker Requirement -AutoGenBench also requires Docker (Desktop or Engine). **It will not run in GitHub codespaces**, unless you opt for native execution (with is strongly discouraged). To install Docker Desktop see [https://www.docker.com/products/docker-desktop/](https://www.docker.com/products/docker-desktop/). +AutoGenBench also requires Docker (Desktop or Engine). **It will not run in GitHub codespaces**, unless you opt for native execution (which is strongly discouraged). To install Docker Desktop see [https://www.docker.com/products/docker-desktop/](https://www.docker.com/products/docker-desktop/). If you are working in WSL, you can follow the instructions below to set up your environment: From 3c1ec7108a658feea01228bf49c1de4008d195f1 Mon Sep 17 00:00:00 2001 From: Eric Zhu Date: Thu, 21 Nov 2024 19:24:12 -0500 Subject: [PATCH 02/10] Misc doc fixes (#4300) * Misc doc fixes * Update _console.py --------- Co-authored-by: Jack Gerrits --- .../src/autogen_agentchat/task/_console.py | 12 +++++++----- .../src/autogen_agentchat/task/_terminations.py | 1 + .../agentchat-user-guide/tutorial/teams.ipynb | 6 +++--- 3 files changed, 11 insertions(+), 8 deletions(-) diff --git a/python/packages/autogen-agentchat/src/autogen_agentchat/task/_console.py b/python/packages/autogen-agentchat/src/autogen_agentchat/task/_console.py index 6b5849e4cf4b..9899366cdc07 100644 --- a/python/packages/autogen-agentchat/src/autogen_agentchat/task/_console.py +++ b/python/packages/autogen-agentchat/src/autogen_agentchat/task/_console.py @@ -27,16 +27,18 @@ async def Console( no_inline_images: bool = False, ) -> T: """ - Consume the stream from :meth:`~autogen_agentchat.base.Team.run_stream` - or :meth:`~autogen_agentchat.base.ChatAgent.on_messages_stream` - print the messages to the console and return the last processed TaskResult or Response. + Consumes the message stream from :meth:`~autogen_agentchat.base.TaskRunner.run_stream` + or :meth:`~autogen_agentchat.base.ChatAgent.on_messages_stream` and renders the messages to the console. + Returns the last processed TaskResult or Response. Args: - stream (AsyncGenerator[AgentMessage | TaskResult, None] | AsyncGenerator[AgentMessage | Response, None]): Stream to render + stream (AsyncGenerator[AgentMessage | TaskResult, None] | AsyncGenerator[AgentMessage | Response, None]): Message stream to render. + This can be from :meth:`~autogen_agentchat.base.TaskRunner.run_stream` or :meth:`~autogen_agentchat.base.ChatAgent.on_messages_stream`. no_inline_images (bool, optional): If terminal is iTerm2 will render images inline. Use this to disable this behavior. Defaults to False. Returns: - last_processed: The last processed TaskResult or Response. + last_processed: A :class:`~autogen_agentchat.base.TaskResult` if the stream is from :meth:`~autogen_agentchat.base.TaskRunner.run_stream` + or a :class:`~autogen_agentchat.base.Response` if the stream is from :meth:`~autogen_agentchat.base.ChatAgent.on_messages_stream`. """ render_image_iterm = _is_running_in_iterm() and _is_output_a_tty() and not no_inline_images start_time = time.time() diff --git a/python/packages/autogen-agentchat/src/autogen_agentchat/task/_terminations.py b/python/packages/autogen-agentchat/src/autogen_agentchat/task/_terminations.py index 31b8d9d3eec0..f8d79cef2850 100644 --- a/python/packages/autogen-agentchat/src/autogen_agentchat/task/_terminations.py +++ b/python/packages/autogen-agentchat/src/autogen_agentchat/task/_terminations.py @@ -237,6 +237,7 @@ def terminated(self) -> bool: return self._terminated def set(self) -> None: + """Set the termination condition to terminated.""" self._setted = True async def __call__(self, messages: Sequence[AgentMessage]) -> StopMessage | None: diff --git a/python/packages/autogen-core/docs/src/user-guide/agentchat-user-guide/tutorial/teams.ipynb b/python/packages/autogen-core/docs/src/user-guide/agentchat-user-guide/tutorial/teams.ipynb index 49340ef11b05..5c0b257dfec2 100644 --- a/python/packages/autogen-core/docs/src/user-guide/agentchat-user-guide/tutorial/teams.ipynb +++ b/python/packages/autogen-core/docs/src/user-guide/agentchat-user-guide/tutorial/teams.ipynb @@ -28,8 +28,8 @@ "\n", "At a high-level, a team API consists of the following methods:\n", "\n", - "- {py:meth}`~autogen_agentchat.base.TaskRunner.run`: To process a task, which can be a {py:class}`str`, {py:class}`~autogen_agentchat.messages.TextMessage`, or {py:class}`~autogen_agentchat.messages.MultiModalMessage`, and returns {py:class}`~autogen_agentchat.base.TaskResult`. The task can also be `None` to resume processing the previous task if the team has not been reset.\n", - "- {py:meth}`~autogen_agentchat.base.TaskRunner.run_stream`: Same as {py:meth}`~autogen_agentchat.base.TaskRunner.run`, but returns a async generator of messages and the final task result.\n", + "- {py:meth}`~autogen_agentchat.base.TaskRunner.run`: Process a task, which can be a {py:class}`str`, {py:class}`~autogen_agentchat.messages.TextMessage`, {py:class}`~autogen_agentchat.messages.MultiModalMessage`, or {py:class}`~autogen_agentchat.messages.HandoffMessage`, and returns {py:class}`~autogen_agentchat.base.TaskResult`. The task can also be `None` to resume processing the previous task if the team has not been reset.\n", + "- {py:meth}`~autogen_agentchat.base.TaskRunner.run_stream`: Similar to {py:meth}`~autogen_agentchat.base.TaskRunner.run`, but it returns an async generator of messages and the final task result.\n", "- {py:meth}`~autogen_agentchat.base.Team.reset`: To reset the team state if the next task is not related to the previous task. Otherwise, the team can utilize the context from the previous task to process the next one.\n", "\n", "In this section, we will be using the\n", @@ -782,7 +782,7 @@ "name": "python", "nbconvert_exporter": "python", "pygments_lexer": "ipython3", - "version": "3.11.5" + "version": "3.12.6" } }, "nbformat": 4, From eb67e4ac93ea85621e6a604c95d38d47104a73b8 Mon Sep 17 00:00:00 2001 From: Ryan Sweet Date: Thu, 21 Nov 2024 16:31:13 -0800 Subject: [PATCH 03/10] add appsettings.Development.json to gitignore (#4303) Co-authored-by: Jack Gerrits --- dotnet/.gitignore | 1 + 1 file changed, 1 insertion(+) diff --git a/dotnet/.gitignore b/dotnet/.gitignore index 2fc32d9ac7e4..62205af71a07 100644 --- a/dotnet/.gitignore +++ b/dotnet/.gitignore @@ -82,6 +82,7 @@ BenchmarkDotNet.Artifacts/ project.lock.json project.fragment.lock.json artifacts/ +appsettings.Development.json # Tye .tye/ From 97fd6cc1e091fe6c0e5226e747b5b0fb6ecce2d6 Mon Sep 17 00:00:00 2001 From: Ryan Sweet Date: Fri, 22 Nov 2024 05:57:11 -0800 Subject: [PATCH 04/10] improve subscriptions (#4304) --- .../Agents/Services/Orleans/SubscriptionsGrain.cs | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-) diff --git a/dotnet/src/Microsoft.AutoGen/Agents/Services/Orleans/SubscriptionsGrain.cs b/dotnet/src/Microsoft.AutoGen/Agents/Services/Orleans/SubscriptionsGrain.cs index 905dc8e914ac..682073f0b97c 100644 --- a/dotnet/src/Microsoft.AutoGen/Agents/Services/Orleans/SubscriptionsGrain.cs +++ b/dotnet/src/Microsoft.AutoGen/Agents/Services/Orleans/SubscriptionsGrain.cs @@ -6,8 +6,13 @@ namespace Microsoft.AutoGen.Agents; internal sealed class SubscriptionsGrain([PersistentState("state", "PubSubStore")] IPersistentState state) : Grain, ISubscriptionsGrain { private readonly Dictionary> _subscriptions = new(); - public ValueTask>> GetSubscriptions(string agentType) + public ValueTask>> GetSubscriptions(string? agentType = null) { + //if agentType is null, return all subscriptions else filter on agentType + if (agentType != null) + { + return new ValueTask>>(_subscriptions.Where(x => x.Value.Contains(agentType)).ToDictionary(x => x.Key, x => x.Value)); + } return new ValueTask>>(_subscriptions); } public ValueTask Subscribe(string agentType, string topic) From 232068a245043aa7b62b2a93beb66b572b8da1e2 Mon Sep 17 00:00:00 2001 From: Gerardo Moreno Date: Fri, 22 Nov 2024 06:05:52 -0800 Subject: [PATCH 05/10] Add system msg when calling inside the assistant tool loop (#4308) (#4309) --- .../src/autogen_agentchat/agents/_assistant_agent.py | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/python/packages/autogen-agentchat/src/autogen_agentchat/agents/_assistant_agent.py b/python/packages/autogen-agentchat/src/autogen_agentchat/agents/_assistant_agent.py index 7e502498f9b6..cb1eff8d6f6e 100644 --- a/python/packages/autogen-agentchat/src/autogen_agentchat/agents/_assistant_agent.py +++ b/python/packages/autogen-agentchat/src/autogen_agentchat/agents/_assistant_agent.py @@ -279,8 +279,9 @@ async def on_messages_stream( return # Generate an inference result based on the current model context. + llm_messages = self._system_messages + self._model_context result = await self._model_client.create( - self._model_context, tools=self._tools + self._handoff_tools, cancellation_token=cancellation_token + llm_messages, tools=self._tools + self._handoff_tools, cancellation_token=cancellation_token ) self._model_context.append(AssistantMessage(content=result.content, source=self.name)) From ac53961bc8b6545824dbe900d497e72b4079e8d1 Mon Sep 17 00:00:00 2001 From: Leonardo Pinheiro Date: Sat, 23 Nov 2024 02:29:39 +1000 Subject: [PATCH 06/10] Delete autogen-ext refactor deprecations (#4305) * delete files and update dependencies * add explicit config exports * ignore mypy error on nb --------- Co-authored-by: Leonardo Pinheiro Co-authored-by: Jack Gerrits --- .../tutorial/termination.ipynb | 2 +- .../cookbook/local-llms-ollama-litellm.ipynb | 8 +- .../design-patterns/mixture-of-agents.ipynb | 3 +- .../design-patterns/multi-agent-debate.ipynb | 4 +- .../core-user-guide/framework/tools.ipynb | 2 +- .../autogen-core/samples/common/utils.py | 3 +- .../samples/distributed-group-chat/_types.py | 2 +- .../samples/distributed-group-chat/_utils.py | 2 +- .../components/models/__init__.py | 30 - .../components/models/_openai_client.py | 901 ------------------ .../components/models/config/__init__.py | 52 - .../autogen-core/tests/test_tool_agent.py | 127 ++- .../packages/autogen-core/tests/test_tools.py | 24 - .../aca_dynamic_sessions/__init__.py | 21 - .../code_executor/docker_executor/__init__.py | 11 - .../src/autogen_ext/models/__init__.py | 9 +- .../models/_openai/config/__init__.py | 3 + .../autogen_ext/tools/langchain/__init__.py | 7 - .../tests/models/test_openai_model_client.py | 51 +- 19 files changed, 131 insertions(+), 1131 deletions(-) delete mode 100644 python/packages/autogen-core/src/autogen_core/components/models/_openai_client.py delete mode 100644 python/packages/autogen-core/src/autogen_core/components/models/config/__init__.py delete mode 100644 python/packages/autogen-ext/src/autogen_ext/code_executor/aca_dynamic_sessions/__init__.py delete mode 100644 python/packages/autogen-ext/src/autogen_ext/code_executor/docker_executor/__init__.py delete mode 100644 python/packages/autogen-ext/src/autogen_ext/tools/langchain/__init__.py diff --git a/python/packages/autogen-core/docs/src/user-guide/agentchat-user-guide/tutorial/termination.ipynb b/python/packages/autogen-core/docs/src/user-guide/agentchat-user-guide/tutorial/termination.ipynb index e10942491286..4a1cfe42cf6c 100644 --- a/python/packages/autogen-core/docs/src/user-guide/agentchat-user-guide/tutorial/termination.ipynb +++ b/python/packages/autogen-core/docs/src/user-guide/agentchat-user-guide/tutorial/termination.ipynb @@ -39,7 +39,7 @@ "from autogen_agentchat.logging import ConsoleLogHandler\n", "from autogen_agentchat.task import MaxMessageTermination, TextMentionTermination\n", "from autogen_agentchat.teams import RoundRobinGroupChat\n", - "from autogen_core.components.models import OpenAIChatCompletionClient\n", + "from autogen_ext.models import OpenAIChatCompletionClient\n", "\n", "logger = logging.getLogger(EVENT_LOGGER_NAME)\n", "logger.addHandler(ConsoleLogHandler())\n", diff --git a/python/packages/autogen-core/docs/src/user-guide/core-user-guide/cookbook/local-llms-ollama-litellm.ipynb b/python/packages/autogen-core/docs/src/user-guide/core-user-guide/cookbook/local-llms-ollama-litellm.ipynb index a90cb440d6ce..80fde2b71017 100644 --- a/python/packages/autogen-core/docs/src/user-guide/core-user-guide/cookbook/local-llms-ollama-litellm.ipynb +++ b/python/packages/autogen-core/docs/src/user-guide/core-user-guide/cookbook/local-llms-ollama-litellm.ipynb @@ -46,10 +46,10 @@ "from autogen_core.components.models import (\n", " AssistantMessage,\n", " ChatCompletionClient,\n", - " OpenAIChatCompletionClient,\n", " SystemMessage,\n", " UserMessage,\n", - ")" + ")\n", + "from autogen_ext.models import OpenAIChatCompletionClient" ] }, { @@ -65,7 +65,7 @@ "metadata": {}, "outputs": [], "source": [ - "def get_model_client() -> OpenAIChatCompletionClient:\n", + "def get_model_client() -> OpenAIChatCompletionClient: # type: ignore\n", " \"Mimic OpenAI API using Local LLM Server.\"\n", " return OpenAIChatCompletionClient(\n", " model=\"gpt-4o\", # Need to use one of the OpenAI models as a placeholder for now.\n", @@ -233,7 +233,7 @@ "name": "python", "nbconvert_exporter": "python", "pygments_lexer": "ipython3", - "version": "3.11.9" + "version": "3.12.7" } }, "nbformat": 4, diff --git a/python/packages/autogen-core/docs/src/user-guide/core-user-guide/design-patterns/mixture-of-agents.ipynb b/python/packages/autogen-core/docs/src/user-guide/core-user-guide/design-patterns/mixture-of-agents.ipynb index dff7d18bd424..61b8b62bc221 100644 --- a/python/packages/autogen-core/docs/src/user-guide/core-user-guide/design-patterns/mixture-of-agents.ipynb +++ b/python/packages/autogen-core/docs/src/user-guide/core-user-guide/design-patterns/mixture-of-agents.ipynb @@ -41,7 +41,8 @@ "from autogen_core.application import SingleThreadedAgentRuntime\n", "from autogen_core.base import AgentId, MessageContext\n", "from autogen_core.components import RoutedAgent, message_handler\n", - "from autogen_core.components.models import ChatCompletionClient, OpenAIChatCompletionClient, SystemMessage, UserMessage" + "from autogen_core.components.models import ChatCompletionClient, SystemMessage, UserMessage\n", + "from autogen_ext.models import OpenAIChatCompletionClient" ] }, { diff --git a/python/packages/autogen-core/docs/src/user-guide/core-user-guide/design-patterns/multi-agent-debate.ipynb b/python/packages/autogen-core/docs/src/user-guide/core-user-guide/design-patterns/multi-agent-debate.ipynb index 3120363dd23b..72b653687915 100644 --- a/python/packages/autogen-core/docs/src/user-guide/core-user-guide/design-patterns/multi-agent-debate.ipynb +++ b/python/packages/autogen-core/docs/src/user-guide/core-user-guide/design-patterns/multi-agent-debate.ipynb @@ -50,10 +50,10 @@ " AssistantMessage,\n", " ChatCompletionClient,\n", " LLMMessage,\n", - " OpenAIChatCompletionClient,\n", " SystemMessage,\n", " UserMessage,\n", - ")" + ")\n", + "from autogen_ext.models import OpenAIChatCompletionClient" ] }, { diff --git a/python/packages/autogen-core/docs/src/user-guide/core-user-guide/framework/tools.ipynb b/python/packages/autogen-core/docs/src/user-guide/core-user-guide/framework/tools.ipynb index 183d878e4c8f..ff24095e8b50 100644 --- a/python/packages/autogen-core/docs/src/user-guide/core-user-guide/framework/tools.ipynb +++ b/python/packages/autogen-core/docs/src/user-guide/core-user-guide/framework/tools.ipynb @@ -161,12 +161,12 @@ "from autogen_core.components.models import (\n", " ChatCompletionClient,\n", " LLMMessage,\n", - " OpenAIChatCompletionClient,\n", " SystemMessage,\n", " UserMessage,\n", ")\n", "from autogen_core.components.tool_agent import ToolAgent, tool_agent_caller_loop\n", "from autogen_core.components.tools import FunctionTool, Tool, ToolSchema\n", + "from autogen_ext.models import OpenAIChatCompletionClient\n", "\n", "\n", "@dataclass\n", diff --git a/python/packages/autogen-core/samples/common/utils.py b/python/packages/autogen-core/samples/common/utils.py index 4e77ac33232e..0765ceec561a 100644 --- a/python/packages/autogen-core/samples/common/utils.py +++ b/python/packages/autogen-core/samples/common/utils.py @@ -3,14 +3,13 @@ from autogen_core.components.models import ( AssistantMessage, - AzureOpenAIChatCompletionClient, ChatCompletionClient, FunctionExecutionResult, FunctionExecutionResultMessage, LLMMessage, - OpenAIChatCompletionClient, UserMessage, ) +from autogen_ext.models import AzureOpenAIChatCompletionClient, OpenAIChatCompletionClient from azure.identity import DefaultAzureCredential, get_bearer_token_provider from typing_extensions import Literal diff --git a/python/packages/autogen-core/samples/distributed-group-chat/_types.py b/python/packages/autogen-core/samples/distributed-group-chat/_types.py index 178446ca8c62..0e05d941c1ff 100644 --- a/python/packages/autogen-core/samples/distributed-group-chat/_types.py +++ b/python/packages/autogen-core/samples/distributed-group-chat/_types.py @@ -4,7 +4,7 @@ from autogen_core.components.models import ( LLMMessage, ) -from autogen_core.components.models.config import AzureOpenAIClientConfiguration +from autogen_ext.models import AzureOpenAIClientConfiguration from pydantic import BaseModel diff --git a/python/packages/autogen-core/samples/distributed-group-chat/_utils.py b/python/packages/autogen-core/samples/distributed-group-chat/_utils.py index 2c4b768e49da..431a94319fc5 100644 --- a/python/packages/autogen-core/samples/distributed-group-chat/_utils.py +++ b/python/packages/autogen-core/samples/distributed-group-chat/_utils.py @@ -5,7 +5,7 @@ import yaml from _types import AppConfig from autogen_core.base import MessageSerializer, try_get_known_serializers_for_type -from autogen_core.components.models.config import AzureOpenAIClientConfiguration +from autogen_ext.models import AzureOpenAIClientConfiguration from azure.identity import DefaultAzureCredential, get_bearer_token_provider diff --git a/python/packages/autogen-core/src/autogen_core/components/models/__init__.py b/python/packages/autogen-core/src/autogen_core/components/models/__init__.py index f57c82289ddc..9b12aa702edd 100644 --- a/python/packages/autogen-core/src/autogen_core/components/models/__init__.py +++ b/python/packages/autogen-core/src/autogen_core/components/models/__init__.py @@ -1,7 +1,3 @@ -import importlib -import warnings -from typing import TYPE_CHECKING, Any - from ._model_client import ChatCompletionClient, ModelCapabilities from ._types import ( AssistantMessage, @@ -17,13 +13,7 @@ UserMessage, ) -if TYPE_CHECKING: - from ._openai_client import AzureOpenAIChatCompletionClient, OpenAIChatCompletionClient - - __all__ = [ - "AzureOpenAIChatCompletionClient", - "OpenAIChatCompletionClient", "ModelCapabilities", "ChatCompletionClient", "SystemMessage", @@ -38,23 +28,3 @@ "TopLogprob", "ChatCompletionTokenLogprob", ] - - -def __getattr__(name: str) -> Any: - deprecated_classes = { - "AzureOpenAIChatCompletionClient": "autogen_ext.models.AzureOpenAIChatCompletionClient", - "OpenAIChatCompletionClient": "autogen_ext.modelsChatCompletionClient", - } - if name in deprecated_classes: - warnings.warn( - f"{name} moved to autogen_ext. " f"Please import it from {deprecated_classes[name]}.", - FutureWarning, - stacklevel=2, - ) - # Dynamically import the class from the current module - module = importlib.import_module("._openai_client", __name__) - attr = getattr(module, name) - # Cache the attribute in the module's global namespace - globals()[name] = attr - return attr - raise AttributeError(f"module {__name__} has no attribute {name}") diff --git a/python/packages/autogen-core/src/autogen_core/components/models/_openai_client.py b/python/packages/autogen-core/src/autogen_core/components/models/_openai_client.py deleted file mode 100644 index 8ce8ddff2cbc..000000000000 --- a/python/packages/autogen-core/src/autogen_core/components/models/_openai_client.py +++ /dev/null @@ -1,901 +0,0 @@ -import asyncio -import inspect -import json -import logging -import math -import re -import warnings -from asyncio import Task -from typing import ( - Any, - AsyncGenerator, - Dict, - List, - Mapping, - Optional, - Sequence, - Set, - Type, - Union, - cast, -) - -import tiktoken -from openai import AsyncAzureOpenAI, AsyncOpenAI -from openai.types.chat import ( - ChatCompletion, - ChatCompletionAssistantMessageParam, - ChatCompletionContentPartParam, - ChatCompletionContentPartTextParam, - ChatCompletionMessageParam, - ChatCompletionMessageToolCallParam, - ChatCompletionRole, - ChatCompletionSystemMessageParam, - ChatCompletionToolMessageParam, - ChatCompletionToolParam, - ChatCompletionUserMessageParam, - ParsedChatCompletion, - ParsedChoice, - completion_create_params, -) -from openai.types.chat.chat_completion import Choice -from openai.types.chat.chat_completion_chunk import Choice as ChunkChoice -from openai.types.shared_params import FunctionDefinition, FunctionParameters -from pydantic import BaseModel -from typing_extensions import Unpack - -from ...application.logging import EVENT_LOGGER_NAME, TRACE_LOGGER_NAME -from ...application.logging.events import LLMCallEvent -from ...base import CancellationToken -from .. import ( - FunctionCall, - Image, -) -from ..tools import Tool, ToolSchema -from . import _model_info -from ._model_client import ChatCompletionClient, ModelCapabilities -from ._types import ( - AssistantMessage, - ChatCompletionTokenLogprob, - CreateResult, - FunctionExecutionResultMessage, - LLMMessage, - RequestUsage, - SystemMessage, - TopLogprob, - UserMessage, -) -from .config import AzureOpenAIClientConfiguration, OpenAIClientConfiguration - -logger = logging.getLogger(EVENT_LOGGER_NAME) -trace_logger = logging.getLogger(TRACE_LOGGER_NAME) - -openai_init_kwargs = set(inspect.getfullargspec(AsyncOpenAI.__init__).kwonlyargs) -aopenai_init_kwargs = set(inspect.getfullargspec(AsyncAzureOpenAI.__init__).kwonlyargs) - -create_kwargs = set(completion_create_params.CompletionCreateParamsBase.__annotations__.keys()) | set( - ("timeout", "stream") -) -# Only single choice allowed -disallowed_create_args = set(["stream", "messages", "function_call", "functions", "n"]) -required_create_args: Set[str] = set(["model"]) - - -def _azure_openai_client_from_config(config: Mapping[str, Any]) -> AsyncAzureOpenAI: - # Take a copy - copied_config = dict(config).copy() - - # Do some fixups - copied_config["azure_deployment"] = copied_config.get("azure_deployment", config.get("model")) - if copied_config["azure_deployment"] is not None: - copied_config["azure_deployment"] = copied_config["azure_deployment"].replace(".", "") - copied_config["azure_endpoint"] = copied_config.get("azure_endpoint", copied_config.pop("base_url", None)) - - # Shave down the config to just the AzureOpenAIChatCompletionClient kwargs - azure_config = {k: v for k, v in copied_config.items() if k in aopenai_init_kwargs} - return AsyncAzureOpenAI(**azure_config) - - -def _openai_client_from_config(config: Mapping[str, Any]) -> AsyncOpenAI: - # Shave down the config to just the OpenAI kwargs - openai_config = {k: v for k, v in config.items() if k in openai_init_kwargs} - return AsyncOpenAI(**openai_config) - - -def _create_args_from_config(config: Mapping[str, Any]) -> Dict[str, Any]: - create_args = {k: v for k, v in config.items() if k in create_kwargs} - create_args_keys = set(create_args.keys()) - if not required_create_args.issubset(create_args_keys): - raise ValueError(f"Required create args are missing: {required_create_args - create_args_keys}") - if disallowed_create_args.intersection(create_args_keys): - raise ValueError(f"Disallowed create args are present: {disallowed_create_args.intersection(create_args_keys)}") - return create_args - - -# TODO check types -# oai_system_message_schema = type2schema(ChatCompletionSystemMessageParam) -# oai_user_message_schema = type2schema(ChatCompletionUserMessageParam) -# oai_assistant_message_schema = type2schema(ChatCompletionAssistantMessageParam) -# oai_tool_message_schema = type2schema(ChatCompletionToolMessageParam) - - -def type_to_role(message: LLMMessage) -> ChatCompletionRole: - if isinstance(message, SystemMessage): - return "system" - elif isinstance(message, UserMessage): - return "user" - elif isinstance(message, AssistantMessage): - return "assistant" - else: - return "tool" - - -def user_message_to_oai(message: UserMessage) -> ChatCompletionUserMessageParam: - assert_valid_name(message.source) - if isinstance(message.content, str): - return ChatCompletionUserMessageParam( - content=message.content, - role="user", - name=message.source, - ) - else: - parts: List[ChatCompletionContentPartParam] = [] - for part in message.content: - if isinstance(part, str): - oai_part = ChatCompletionContentPartTextParam( - text=part, - type="text", - ) - parts.append(oai_part) - elif isinstance(part, Image): - # TODO: support url based images - # TODO: support specifying details - parts.append(part.to_openai_format()) - else: - raise ValueError(f"Unknown content type: {part}") - return ChatCompletionUserMessageParam( - content=parts, - role="user", - name=message.source, - ) - - -def system_message_to_oai(message: SystemMessage) -> ChatCompletionSystemMessageParam: - return ChatCompletionSystemMessageParam( - content=message.content, - role="system", - ) - - -def func_call_to_oai(message: FunctionCall) -> ChatCompletionMessageToolCallParam: - return ChatCompletionMessageToolCallParam( - id=message.id, - function={ - "arguments": message.arguments, - "name": message.name, - }, - type="function", - ) - - -def tool_message_to_oai( - message: FunctionExecutionResultMessage, -) -> Sequence[ChatCompletionToolMessageParam]: - return [ - ChatCompletionToolMessageParam(content=x.content, role="tool", tool_call_id=x.call_id) for x in message.content - ] - - -def assistant_message_to_oai( - message: AssistantMessage, -) -> ChatCompletionAssistantMessageParam: - assert_valid_name(message.source) - if isinstance(message.content, list): - return ChatCompletionAssistantMessageParam( - tool_calls=[func_call_to_oai(x) for x in message.content], - role="assistant", - name=message.source, - ) - else: - return ChatCompletionAssistantMessageParam( - content=message.content, - role="assistant", - name=message.source, - ) - - -def to_oai_type(message: LLMMessage) -> Sequence[ChatCompletionMessageParam]: - if isinstance(message, SystemMessage): - return [system_message_to_oai(message)] - elif isinstance(message, UserMessage): - return [user_message_to_oai(message)] - elif isinstance(message, AssistantMessage): - return [assistant_message_to_oai(message)] - else: - return tool_message_to_oai(message) - - -def calculate_vision_tokens(image: Image, detail: str = "auto") -> int: - MAX_LONG_EDGE = 2048 - BASE_TOKEN_COUNT = 85 - TOKENS_PER_TILE = 170 - MAX_SHORT_EDGE = 768 - TILE_SIZE = 512 - - if detail == "low": - return BASE_TOKEN_COUNT - - width, height = image.image.size - - # Scale down to fit within a MAX_LONG_EDGE x MAX_LONG_EDGE square if necessary - - if width > MAX_LONG_EDGE or height > MAX_LONG_EDGE: - aspect_ratio = width / height - if aspect_ratio > 1: - # Width is greater than height - width = MAX_LONG_EDGE - height = int(MAX_LONG_EDGE / aspect_ratio) - else: - # Height is greater than or equal to width - height = MAX_LONG_EDGE - width = int(MAX_LONG_EDGE * aspect_ratio) - - # Resize such that the shortest side is MAX_SHORT_EDGE if both dimensions exceed MAX_SHORT_EDGE - aspect_ratio = width / height - if width > MAX_SHORT_EDGE and height > MAX_SHORT_EDGE: - if aspect_ratio > 1: - # Width is greater than height - height = MAX_SHORT_EDGE - width = int(MAX_SHORT_EDGE * aspect_ratio) - else: - # Height is greater than or equal to width - width = MAX_SHORT_EDGE - height = int(MAX_SHORT_EDGE / aspect_ratio) - - # Calculate the number of tiles based on TILE_SIZE - - tiles_width = math.ceil(width / TILE_SIZE) - tiles_height = math.ceil(height / TILE_SIZE) - total_tiles = tiles_width * tiles_height - # Calculate the total tokens based on the number of tiles and the base token count - - total_tokens = BASE_TOKEN_COUNT + TOKENS_PER_TILE * total_tiles - - return total_tokens - - -def _add_usage(usage1: RequestUsage, usage2: RequestUsage) -> RequestUsage: - return RequestUsage( - prompt_tokens=usage1.prompt_tokens + usage2.prompt_tokens, - completion_tokens=usage1.completion_tokens + usage2.completion_tokens, - ) - - -def convert_tools( - tools: Sequence[Tool | ToolSchema], -) -> List[ChatCompletionToolParam]: - result: List[ChatCompletionToolParam] = [] - for tool in tools: - if isinstance(tool, Tool): - tool_schema = tool.schema - else: - assert isinstance(tool, dict) - tool_schema = tool - - result.append( - ChatCompletionToolParam( - type="function", - function=FunctionDefinition( - name=tool_schema["name"], - description=(tool_schema["description"] if "description" in tool_schema else ""), - parameters=( - cast(FunctionParameters, tool_schema["parameters"]) if "parameters" in tool_schema else {} - ), - ), - ) - ) - # Check if all tools have valid names. - for tool_param in result: - assert_valid_name(tool_param["function"]["name"]) - return result - - -def normalize_name(name: str) -> str: - """ - LLMs sometimes ask functions while ignoring their own format requirements, this function should be used to replace invalid characters with "_". - - Prefer _assert_valid_name for validating user configuration or input - """ - return re.sub(r"[^a-zA-Z0-9_-]", "_", name)[:64] - - -def assert_valid_name(name: str) -> str: - """ - Ensure that configured names are valid, raises ValueError if not. - - For munging LLM responses use _normalize_name to ensure LLM specified names don't break the API. - """ - if not re.match(r"^[a-zA-Z0-9_-]+$", name): - raise ValueError(f"Invalid name: {name}. Only letters, numbers, '_' and '-' are allowed.") - if len(name) > 64: - raise ValueError(f"Invalid name: {name}. Name must be less than 64 characters.") - return name - - -class BaseOpenAIChatCompletionClient(ChatCompletionClient): - def __init__( - self, - client: Union[AsyncOpenAI, AsyncAzureOpenAI], - create_args: Dict[str, Any], - model_capabilities: Optional[ModelCapabilities] = None, - ): - self._client = client - if model_capabilities is None and isinstance(client, AsyncAzureOpenAI): - raise ValueError("AzureOpenAIChatCompletionClient requires explicit model capabilities") - elif model_capabilities is None: - self._model_capabilities = _model_info.get_capabilities(create_args["model"]) - else: - self._model_capabilities = model_capabilities - - self._resolved_model: Optional[str] = None - if "model" in create_args: - self._resolved_model = _model_info.resolve_model(create_args["model"]) - - if ( - "response_format" in create_args - and create_args["response_format"]["type"] == "json_object" - and not self._model_capabilities["json_output"] - ): - raise ValueError("Model does not support JSON output") - - self._create_args = create_args - self._total_usage = RequestUsage(prompt_tokens=0, completion_tokens=0) - self._actual_usage = RequestUsage(prompt_tokens=0, completion_tokens=0) - - @classmethod - def create_from_config(cls, config: Dict[str, Any]) -> ChatCompletionClient: - return OpenAIChatCompletionClient(**config) - - async def create( - self, - messages: Sequence[LLMMessage], - tools: Sequence[Tool | ToolSchema] = [], - json_output: Optional[bool] = None, - extra_create_args: Mapping[str, Any] = {}, - cancellation_token: Optional[CancellationToken] = None, - ) -> CreateResult: - # Make sure all extra_create_args are valid - extra_create_args_keys = set(extra_create_args.keys()) - if not create_kwargs.issuperset(extra_create_args_keys): - raise ValueError(f"Extra create args are invalid: {extra_create_args_keys - create_kwargs}") - - # Copy the create args and overwrite anything in extra_create_args - create_args = self._create_args.copy() - create_args.update(extra_create_args) - - # Declare use_beta_client - use_beta_client: bool = False - response_format_value: Optional[Type[BaseModel]] = None - - if "response_format" in create_args: - value = create_args["response_format"] - # If value is a Pydantic model class, use the beta client - if isinstance(value, type) and issubclass(value, BaseModel): - response_format_value = value - use_beta_client = True - else: - # response_format_value is not a Pydantic model class - use_beta_client = False - response_format_value = None - - # Remove 'response_format' from create_args to prevent passing it twice - create_args_no_response_format = {k: v for k, v in create_args.items() if k != "response_format"} - - # TODO: allow custom handling. - # For now we raise an error if images are present and vision is not supported - if self.capabilities["vision"] is False: - for message in messages: - if isinstance(message, UserMessage): - if isinstance(message.content, list) and any(isinstance(x, Image) for x in message.content): - raise ValueError("Model does not support vision and image was provided") - - if json_output is not None: - if self.capabilities["json_output"] is False and json_output is True: - raise ValueError("Model does not support JSON output") - - if json_output is True: - create_args["response_format"] = {"type": "json_object"} - else: - create_args["response_format"] = {"type": "text"} - - if self.capabilities["json_output"] is False and json_output is True: - raise ValueError("Model does not support JSON output") - - oai_messages_nested = [to_oai_type(m) for m in messages] - oai_messages = [item for sublist in oai_messages_nested for item in sublist] - - if self.capabilities["function_calling"] is False and len(tools) > 0: - raise ValueError("Model does not support function calling") - future: Union[Task[ParsedChatCompletion[BaseModel]], Task[ChatCompletion]] - if len(tools) > 0: - converted_tools = convert_tools(tools) - if use_beta_client: - # Pass response_format_value if it's not None - if response_format_value is not None: - future = asyncio.ensure_future( - self._client.beta.chat.completions.parse( - messages=oai_messages, - tools=converted_tools, - response_format=response_format_value, - **create_args_no_response_format, - ) - ) - else: - future = asyncio.ensure_future( - self._client.beta.chat.completions.parse( - messages=oai_messages, - tools=converted_tools, - **create_args_no_response_format, - ) - ) - else: - future = asyncio.ensure_future( - self._client.chat.completions.create( - messages=oai_messages, - stream=False, - tools=converted_tools, - **create_args, - ) - ) - else: - if use_beta_client: - if response_format_value is not None: - future = asyncio.ensure_future( - self._client.beta.chat.completions.parse( - messages=oai_messages, - response_format=response_format_value, - **create_args_no_response_format, - ) - ) - else: - future = asyncio.ensure_future( - self._client.beta.chat.completions.parse( - messages=oai_messages, - **create_args_no_response_format, - ) - ) - else: - future = asyncio.ensure_future( - self._client.chat.completions.create( - messages=oai_messages, - stream=False, - **create_args, - ) - ) - - if cancellation_token is not None: - cancellation_token.link_future(future) - result: Union[ParsedChatCompletion[BaseModel], ChatCompletion] = await future - if use_beta_client: - result = cast(ParsedChatCompletion[Any], result) - - if result.usage is not None: - logger.info( - LLMCallEvent( - prompt_tokens=result.usage.prompt_tokens, - completion_tokens=result.usage.completion_tokens, - ) - ) - - usage = RequestUsage( - # TODO backup token counting - prompt_tokens=result.usage.prompt_tokens if result.usage is not None else 0, - completion_tokens=(result.usage.completion_tokens if result.usage is not None else 0), - ) - - if self._resolved_model is not None: - if self._resolved_model != result.model: - warnings.warn( - f"Resolved model mismatch: {self._resolved_model} != {result.model}. Model mapping may be incorrect.", - stacklevel=2, - ) - - # Limited to a single choice currently. - choice: Union[ParsedChoice[Any], ParsedChoice[BaseModel], Choice] = result.choices[0] - if choice.finish_reason == "function_call": - raise ValueError("Function calls are not supported in this context") - - content: Union[str, List[FunctionCall]] - if choice.finish_reason == "tool_calls": - assert choice.message.tool_calls is not None - assert choice.message.function_call is None - - # NOTE: If OAI response type changes, this will need to be updated - content = [ - FunctionCall( - id=x.id, - arguments=x.function.arguments, - name=normalize_name(x.function.name), - ) - for x in choice.message.tool_calls - ] - finish_reason = "function_calls" - else: - finish_reason = choice.finish_reason - content = choice.message.content or "" - logprobs: Optional[List[ChatCompletionTokenLogprob]] = None - if choice.logprobs and choice.logprobs.content: - logprobs = [ - ChatCompletionTokenLogprob( - token=x.token, - logprob=x.logprob, - top_logprobs=[TopLogprob(logprob=y.logprob, bytes=y.bytes) for y in x.top_logprobs], - bytes=x.bytes, - ) - for x in choice.logprobs.content - ] - response = CreateResult( - finish_reason=finish_reason, # type: ignore - content=content, - usage=usage, - cached=False, - logprobs=logprobs, - ) - - _add_usage(self._actual_usage, usage) - _add_usage(self._total_usage, usage) - - # TODO - why is this cast needed? - return response - - async def create_stream( - self, - messages: Sequence[LLMMessage], - tools: Sequence[Tool | ToolSchema] = [], - json_output: Optional[bool] = None, - extra_create_args: Mapping[str, Any] = {}, - cancellation_token: Optional[CancellationToken] = None, - ) -> AsyncGenerator[Union[str, CreateResult], None]: - """ - Creates an AsyncGenerator that will yield a stream of chat completions based on the provided messages and tools. - - Args: - messages (Sequence[LLMMessage]): A sequence of messages to be processed. - tools (Sequence[Tool | ToolSchema], optional): A sequence of tools to be used in the completion. Defaults to `[]`. - json_output (Optional[bool], optional): If True, the output will be in JSON format. Defaults to None. - extra_create_args (Mapping[str, Any], optional): Additional arguments for the creation process. Default to `{}`. - cancellation_token (Optional[CancellationToken], optional): A token to cancel the operation. Defaults to None. - - Yields: - AsyncGenerator[Union[str, CreateResult], None]: A generator yielding the completion results as they are produced. - - In streaming, the default behaviour is not return token usage counts. See: [OpenAI API reference for possible args](https://platform.openai.com/docs/api-reference/chat/create). - However `extra_create_args={"stream_options": {"include_usage": True}}` will (if supported by the accessed API) - return a final chunk with usage set to a RequestUsage object having prompt and completion token counts, - all preceding chunks will have usage as None. See: [stream_options](https://platform.openai.com/docs/api-reference/chat/create#chat-create-stream_options). - - Other examples of OPENAI supported arguments that can be included in `extra_create_args`: - - `temperature` (float): Controls the randomness of the output. Higher values (e.g., 0.8) make the output more random, while lower values (e.g., 0.2) make it more focused and deterministic. - - `max_tokens` (int): The maximum number of tokens to generate in the completion. - - `top_p` (float): An alternative to sampling with temperature, called nucleus sampling, where the model considers the results of the tokens with top_p probability mass. - - `frequency_penalty` (float): A value between -2.0 and 2.0 that penalizes new tokens based on their existing frequency in the text so far, decreasing the likelihood of repeated phrases. - - `presence_penalty` (float): A value between -2.0 and 2.0 that penalizes new tokens based on whether they appear in the text so far, encouraging the model to talk about new topics. - """ - # Make sure all extra_create_args are valid - extra_create_args_keys = set(extra_create_args.keys()) - if not create_kwargs.issuperset(extra_create_args_keys): - raise ValueError(f"Extra create args are invalid: {extra_create_args_keys - create_kwargs}") - - # Copy the create args and overwrite anything in extra_create_args - create_args = self._create_args.copy() - create_args.update(extra_create_args) - - oai_messages_nested = [to_oai_type(m) for m in messages] - oai_messages = [item for sublist in oai_messages_nested for item in sublist] - - # TODO: allow custom handling. - # For now we raise an error if images are present and vision is not supported - if self.capabilities["vision"] is False: - for message in messages: - if isinstance(message, UserMessage): - if isinstance(message.content, list) and any(isinstance(x, Image) for x in message.content): - raise ValueError("Model does not support vision and image was provided") - - if json_output is not None: - if self.capabilities["json_output"] is False and json_output is True: - raise ValueError("Model does not support JSON output") - - if json_output is True: - create_args["response_format"] = {"type": "json_object"} - else: - create_args["response_format"] = {"type": "text"} - - if len(tools) > 0: - converted_tools = convert_tools(tools) - stream_future = asyncio.ensure_future( - self._client.chat.completions.create( - messages=oai_messages, - stream=True, - tools=converted_tools, - **create_args, - ) - ) - else: - stream_future = asyncio.ensure_future( - self._client.chat.completions.create(messages=oai_messages, stream=True, **create_args) - ) - if cancellation_token is not None: - cancellation_token.link_future(stream_future) - stream = await stream_future - choice: Union[ParsedChoice[Any], ParsedChoice[BaseModel], ChunkChoice] = cast(ChunkChoice, None) - chunk = None - stop_reason = None - maybe_model = None - content_deltas: List[str] = [] - full_tool_calls: Dict[int, FunctionCall] = {} - completion_tokens = 0 - logprobs: Optional[List[ChatCompletionTokenLogprob]] = None - while True: - try: - chunk_future = asyncio.ensure_future(anext(stream)) - if cancellation_token is not None: - cancellation_token.link_future(chunk_future) - chunk = await chunk_future - - # to process usage chunk in streaming situations - # add stream_options={"include_usage": True} in the initialization of OpenAIChatCompletionClient(...) - # However the different api's - # OPENAI api usage chunk produces no choices so need to check if there is a choice - # liteLLM api usage chunk does produce choices - choice = ( - chunk.choices[0] - if len(chunk.choices) > 0 - else choice - if chunk.usage is not None and stop_reason is not None - else cast(ChunkChoice, None) - ) - - # for liteLLM chunk usage, do the following hack keeping the pervious chunk.stop_reason (if set). - # set the stop_reason for the usage chunk to the prior stop_reason - stop_reason = choice.finish_reason if chunk.usage is None and stop_reason is None else stop_reason - maybe_model = chunk.model - # First try get content - if choice.delta.content is not None: - content_deltas.append(choice.delta.content) - if len(choice.delta.content) > 0: - yield choice.delta.content - continue - - # Otherwise, get tool calls - if choice.delta.tool_calls is not None: - for tool_call_chunk in choice.delta.tool_calls: - idx = tool_call_chunk.index - if idx not in full_tool_calls: - # We ignore the type hint here because we want to fill in type when the delta provides it - full_tool_calls[idx] = FunctionCall(id="", arguments="", name="") - - if tool_call_chunk.id is not None: - full_tool_calls[idx].id += tool_call_chunk.id - - if tool_call_chunk.function is not None: - if tool_call_chunk.function.name is not None: - full_tool_calls[idx].name += tool_call_chunk.function.name - if tool_call_chunk.function.arguments is not None: - full_tool_calls[idx].arguments += tool_call_chunk.function.arguments - if choice.logprobs and choice.logprobs.content: - logprobs = [ - ChatCompletionTokenLogprob( - token=x.token, - logprob=x.logprob, - top_logprobs=[TopLogprob(logprob=y.logprob, bytes=y.bytes) for y in x.top_logprobs], - bytes=x.bytes, - ) - for x in choice.logprobs.content - ] - - except StopAsyncIteration: - break - - model = maybe_model or create_args["model"] - model = model.replace("gpt-35", "gpt-3.5") # hack for Azure API - - if chunk and chunk.usage: - prompt_tokens = chunk.usage.prompt_tokens - else: - prompt_tokens = 0 - - if stop_reason is None: - raise ValueError("No stop reason found") - - content: Union[str, List[FunctionCall]] - if len(content_deltas) > 1: - content = "".join(content_deltas) - if chunk and chunk.usage: - completion_tokens = chunk.usage.completion_tokens - else: - completion_tokens = 0 - else: - completion_tokens = 0 - # TODO: fix assumption that dict values were added in order and actually order by int index - # for tool_call in full_tool_calls.values(): - # # value = json.dumps(tool_call) - # # completion_tokens += count_token(value, model=model) - # completion_tokens += 0 - content = list(full_tool_calls.values()) - - usage = RequestUsage( - prompt_tokens=prompt_tokens, - completion_tokens=completion_tokens, - ) - if stop_reason == "function_call": - raise ValueError("Function calls are not supported in this context") - if stop_reason == "tool_calls": - stop_reason = "function_calls" - - result = CreateResult( - finish_reason=stop_reason, # type: ignore - content=content, - usage=usage, - cached=False, - logprobs=logprobs, - ) - - _add_usage(self._actual_usage, usage) - _add_usage(self._total_usage, usage) - - yield result - - def actual_usage(self) -> RequestUsage: - return self._actual_usage - - def total_usage(self) -> RequestUsage: - return self._total_usage - - def count_tokens(self, messages: Sequence[LLMMessage], tools: Sequence[Tool | ToolSchema] = []) -> int: - model = self._create_args["model"] - try: - encoding = tiktoken.encoding_for_model(model) - except KeyError: - trace_logger.warning(f"Model {model} not found. Using cl100k_base encoding.") - encoding = tiktoken.get_encoding("cl100k_base") - tokens_per_message = 3 - tokens_per_name = 1 - num_tokens = 0 - - # Message tokens. - for message in messages: - num_tokens += tokens_per_message - oai_message = to_oai_type(message) - for oai_message_part in oai_message: - for key, value in oai_message_part.items(): - if value is None: - continue - - if isinstance(message, UserMessage) and isinstance(value, list): - typed_message_value = cast(List[ChatCompletionContentPartParam], value) - - assert len(typed_message_value) == len( - message.content - ), "Mismatch in message content and typed message value" - - # We need image properties that are only in the original message - for part, content_part in zip(typed_message_value, message.content, strict=False): - if isinstance(content_part, Image): - # TODO: add detail parameter - num_tokens += calculate_vision_tokens(content_part) - elif isinstance(part, str): - num_tokens += len(encoding.encode(part)) - else: - try: - serialized_part = json.dumps(part) - num_tokens += len(encoding.encode(serialized_part)) - except TypeError: - trace_logger.warning(f"Could not convert {part} to string, skipping.") - else: - if not isinstance(value, str): - try: - value = json.dumps(value) - except TypeError: - trace_logger.warning(f"Could not convert {value} to string, skipping.") - continue - num_tokens += len(encoding.encode(value)) - if key == "name": - num_tokens += tokens_per_name - num_tokens += 3 # every reply is primed with <|start|>assistant<|message|> - - # Tool tokens. - oai_tools = convert_tools(tools) - for tool in oai_tools: - function = tool["function"] - tool_tokens = len(encoding.encode(function["name"])) - if "description" in function: - tool_tokens += len(encoding.encode(function["description"])) - tool_tokens -= 2 - if "parameters" in function: - parameters = function["parameters"] - if "properties" in parameters: - assert isinstance(parameters["properties"], dict) - for propertiesKey in parameters["properties"]: # pyright: ignore - assert isinstance(propertiesKey, str) - tool_tokens += len(encoding.encode(propertiesKey)) - v = parameters["properties"][propertiesKey] # pyright: ignore - for field in v: # pyright: ignore - if field == "type": - tool_tokens += 2 - tool_tokens += len(encoding.encode(v["type"])) # pyright: ignore - elif field == "description": - tool_tokens += 2 - tool_tokens += len(encoding.encode(v["description"])) # pyright: ignore - elif field == "enum": - tool_tokens -= 3 - for o in v["enum"]: # pyright: ignore - tool_tokens += 3 - tool_tokens += len(encoding.encode(o)) # pyright: ignore - else: - trace_logger.warning(f"Not supported field {field}") - tool_tokens += 11 - if len(parameters["properties"]) == 0: # pyright: ignore - tool_tokens -= 2 - num_tokens += tool_tokens - num_tokens += 12 - return num_tokens - - def remaining_tokens(self, messages: Sequence[LLMMessage], tools: Sequence[Tool | ToolSchema] = []) -> int: - token_limit = _model_info.get_token_limit(self._create_args["model"]) - return token_limit - self.count_tokens(messages, tools) - - @property - def capabilities(self) -> ModelCapabilities: - return self._model_capabilities - - -class OpenAIChatCompletionClient(BaseOpenAIChatCompletionClient): - def __init__(self, **kwargs: Unpack[OpenAIClientConfiguration]): - if "model" not in kwargs: - raise ValueError("model is required for OpenAIChatCompletionClient") - - model_capabilities: Optional[ModelCapabilities] = None - copied_args = dict(kwargs).copy() - if "model_capabilities" in kwargs: - model_capabilities = kwargs["model_capabilities"] - del copied_args["model_capabilities"] - - client = _openai_client_from_config(copied_args) - create_args = _create_args_from_config(copied_args) - self._raw_config = copied_args - super().__init__(client, create_args, model_capabilities) - - def __getstate__(self) -> Dict[str, Any]: - state = self.__dict__.copy() - state["_client"] = None - return state - - def __setstate__(self, state: Dict[str, Any]) -> None: - self.__dict__.update(state) - self._client = _openai_client_from_config(state["_raw_config"]) - - -class AzureOpenAIChatCompletionClient(BaseOpenAIChatCompletionClient): - def __init__(self, **kwargs: Unpack[AzureOpenAIClientConfiguration]): - if "model" not in kwargs: - raise ValueError("model is required for OpenAIChatCompletionClient") - - model_capabilities: Optional[ModelCapabilities] = None - copied_args = dict(kwargs).copy() - if "model_capabilities" in kwargs: - model_capabilities = kwargs["model_capabilities"] - del copied_args["model_capabilities"] - - client = _azure_openai_client_from_config(copied_args) - create_args = _create_args_from_config(copied_args) - self._raw_config = copied_args - super().__init__(client, create_args, model_capabilities) - - def __getstate__(self) -> Dict[str, Any]: - state = self.__dict__.copy() - state["_client"] = None - return state - - def __setstate__(self, state: Dict[str, Any]) -> None: - self.__dict__.update(state) - self._client = _azure_openai_client_from_config(state["_raw_config"]) diff --git a/python/packages/autogen-core/src/autogen_core/components/models/config/__init__.py b/python/packages/autogen-core/src/autogen_core/components/models/config/__init__.py deleted file mode 100644 index d1edcf8c62f9..000000000000 --- a/python/packages/autogen-core/src/autogen_core/components/models/config/__init__.py +++ /dev/null @@ -1,52 +0,0 @@ -from typing import Awaitable, Callable, Dict, List, Literal, Optional, Union - -from typing_extensions import Required, TypedDict - -from .._model_client import ModelCapabilities - - -class ResponseFormat(TypedDict): - type: Literal["text", "json_object"] - - -class CreateArguments(TypedDict, total=False): - frequency_penalty: Optional[float] - logit_bias: Optional[Dict[str, int]] - max_tokens: Optional[int] - n: Optional[int] - presence_penalty: Optional[float] - response_format: ResponseFormat - seed: Optional[int] - stop: Union[Optional[str], List[str]] - temperature: Optional[float] - top_p: Optional[float] - user: str - - -AsyncAzureADTokenProvider = Callable[[], Union[str, Awaitable[str]]] - - -class BaseOpenAIClientConfiguration(CreateArguments, total=False): - model: str - api_key: str - timeout: Union[float, None] - max_retries: int - - -# See OpenAI docs for explanation of these parameters -class OpenAIClientConfiguration(BaseOpenAIClientConfiguration, total=False): - organization: str - base_url: str - # Not required - model_capabilities: ModelCapabilities - - -class AzureOpenAIClientConfiguration(BaseOpenAIClientConfiguration, total=False): - # Azure specific - azure_endpoint: Required[str] - azure_deployment: str - api_version: Required[str] - azure_ad_token: str - azure_ad_token_provider: AsyncAzureADTokenProvider - # Must be provided - model_capabilities: Required[ModelCapabilities] diff --git a/python/packages/autogen-core/tests/test_tool_agent.py b/python/packages/autogen-core/tests/test_tool_agent.py index 322fdf6b7941..6184e9c78c83 100644 --- a/python/packages/autogen-core/tests/test_tool_agent.py +++ b/python/packages/autogen-core/tests/test_tool_agent.py @@ -1,6 +1,6 @@ import asyncio import json -from typing import Any, AsyncGenerator, List +from typing import Any, AsyncGenerator, List, Mapping, Optional, Sequence, Union import pytest from autogen_core.application import SingleThreadedAgentRuntime @@ -8,9 +8,13 @@ from autogen_core.components import FunctionCall from autogen_core.components.models import ( AssistantMessage, + ChatCompletionClient, + CreateResult, FunctionExecutionResult, FunctionExecutionResultMessage, - OpenAIChatCompletionClient, + LLMMessage, + ModelCapabilities, + RequestUsage, UserMessage, ) from autogen_core.components.tool_agent import ( @@ -20,13 +24,7 @@ ToolNotFoundException, tool_agent_caller_loop, ) -from autogen_core.components.tools import FunctionTool, Tool -from openai.resources.chat.completions import AsyncCompletions -from openai.types.chat.chat_completion import ChatCompletion, Choice -from openai.types.chat.chat_completion_chunk import ChatCompletionChunk -from openai.types.chat.chat_completion_message import ChatCompletionMessage -from openai.types.chat.chat_completion_message_tool_call import ChatCompletionMessageToolCall, Function -from openai.types.completion_usage import CompletionUsage +from autogen_core.components.tools import FunctionTool, Tool, ToolSchema def _pass_function(input: str) -> str: @@ -42,60 +40,6 @@ async def _async_sleep_function(input: str) -> str: return "pass" -class _MockChatCompletion: - def __init__(self, model: str = "gpt-4o") -> None: - self._saved_chat_completions: List[ChatCompletion] = [ - ChatCompletion( - id="id1", - choices=[ - Choice( - finish_reason="tool_calls", - index=0, - message=ChatCompletionMessage( - content=None, - tool_calls=[ - ChatCompletionMessageToolCall( - id="1", - type="function", - function=Function( - name="pass", - arguments=json.dumps({"input": "pass"}), - ), - ) - ], - role="assistant", - ), - ) - ], - created=0, - model=model, - object="chat.completion", - usage=CompletionUsage(prompt_tokens=0, completion_tokens=0, total_tokens=0), - ), - ChatCompletion( - id="id2", - choices=[ - Choice( - finish_reason="stop", index=0, message=ChatCompletionMessage(content="Hello", role="assistant") - ) - ], - created=0, - model=model, - object="chat.completion", - usage=CompletionUsage(prompt_tokens=0, completion_tokens=0, total_tokens=0), - ), - ] - self._curr_index = 0 - - async def mock_create( - self, *args: Any, **kwargs: Any - ) -> ChatCompletion | AsyncGenerator[ChatCompletionChunk, None]: - await asyncio.sleep(0.1) - completion = self._saved_chat_completions[self._curr_index] - self._curr_index += 1 - return completion - - @pytest.mark.asyncio async def test_tool_agent() -> None: runtime = SingleThreadedAgentRuntime() @@ -144,10 +88,59 @@ async def test_tool_agent() -> None: @pytest.mark.asyncio -async def test_caller_loop(monkeypatch: pytest.MonkeyPatch) -> None: - mock = _MockChatCompletion(model="gpt-4o-2024-05-13") - monkeypatch.setattr(AsyncCompletions, "create", mock.mock_create) - client = OpenAIChatCompletionClient(model="gpt-4o-2024-05-13", api_key="api_key") +async def test_caller_loop() -> None: + class MockChatCompletionClient(ChatCompletionClient): + async def create( + self, + messages: Sequence[LLMMessage], + tools: Sequence[Tool | ToolSchema] = [], + json_output: Optional[bool] = None, + extra_create_args: Mapping[str, Any] = {}, + cancellation_token: Optional[CancellationToken] = None, + ) -> CreateResult: + if len(messages) == 1: + return CreateResult( + content=[FunctionCall(id="1", name="pass", arguments=json.dumps({"input": "test"}))], + finish_reason="stop", + usage=RequestUsage(prompt_tokens=0, completion_tokens=0), + cached=False, + logprobs=None, + ) + return CreateResult( + content="Done", + finish_reason="stop", + usage=RequestUsage(prompt_tokens=0, completion_tokens=0), + cached=False, + logprobs=None, + ) + + def create_stream( + self, + messages: Sequence[LLMMessage], + tools: Sequence[Tool | ToolSchema] = [], + json_output: Optional[bool] = None, + extra_create_args: Mapping[str, Any] = {}, + cancellation_token: Optional[CancellationToken] = None, + ) -> AsyncGenerator[Union[str, CreateResult], None]: + raise NotImplementedError() + + def actual_usage(self) -> RequestUsage: + return RequestUsage(prompt_tokens=0, completion_tokens=0) + + def total_usage(self) -> RequestUsage: + return RequestUsage(prompt_tokens=0, completion_tokens=0) + + def count_tokens(self, messages: Sequence[LLMMessage], tools: Sequence[Tool | ToolSchema] = []) -> int: + return 0 + + def remaining_tokens(self, messages: Sequence[LLMMessage], tools: Sequence[Tool | ToolSchema] = []) -> int: + return 0 + + @property + def capabilities(self) -> ModelCapabilities: + return ModelCapabilities(vision=False, function_calling=True, json_output=False) + + client = MockChatCompletionClient() tools: List[Tool] = [FunctionTool(_pass_function, name="pass", description="Pass function")] runtime = SingleThreadedAgentRuntime() await runtime.register( diff --git a/python/packages/autogen-core/tests/test_tools.py b/python/packages/autogen-core/tests/test_tools.py index 70ec08469706..27a89748c659 100644 --- a/python/packages/autogen-core/tests/test_tools.py +++ b/python/packages/autogen-core/tests/test_tools.py @@ -4,7 +4,6 @@ import pytest from autogen_core.base import CancellationToken from autogen_core.components._function_utils import get_typed_signature -from autogen_core.components.models._openai_client import convert_tools from autogen_core.components.tools import BaseTool, FunctionTool from autogen_core.components.tools._base import ToolSchema from pydantic import BaseModel, Field, model_serializer @@ -323,29 +322,6 @@ def my_function(arg: int) -> int: assert tool.return_value_as_string(result) == "5" -def test_convert_tools_accepts_both_func_tool_and_schema() -> None: - def my_function(arg: str, other: Annotated[int, "int arg"], nonrequired: int = 5) -> MyResult: - return MyResult(result="test") - - tool = FunctionTool(my_function, description="Function tool.") - schema = tool.schema - - converted_tool_schema = convert_tools([tool, schema]) - - assert len(converted_tool_schema) == 2 - assert converted_tool_schema[0] == converted_tool_schema[1] - - -def test_convert_tools_accepts_both_tool_and_schema() -> None: - tool = MyTool() - schema = tool.schema - - converted_tool_schema = convert_tools([tool, schema]) - - assert len(converted_tool_schema) == 2 - assert converted_tool_schema[0] == converted_tool_schema[1] - - @pytest.mark.asyncio async def test_func_tool_return_list() -> None: def my_function() -> List[int]: diff --git a/python/packages/autogen-ext/src/autogen_ext/code_executor/aca_dynamic_sessions/__init__.py b/python/packages/autogen-ext/src/autogen_ext/code_executor/aca_dynamic_sessions/__init__.py deleted file mode 100644 index 009997c41abc..000000000000 --- a/python/packages/autogen-ext/src/autogen_ext/code_executor/aca_dynamic_sessions/__init__.py +++ /dev/null @@ -1,21 +0,0 @@ -import warnings -from typing import Any - -from ...code_executors import ACADynamicSessionsCodeExecutor - - -class AzureContainerCodeExecutor(ACADynamicSessionsCodeExecutor): - """AzureContainerCodeExecutor has been renamed and moved to autogen_ext.code_executors.ACADynamicSessionsCodeExecutor""" - - def __init__(self, *args: Any, **kwargs: Any) -> None: - warnings.warn( - "AzureContainerCodeExecutor has been renamed and moved to autogen_ext.code_executors.ACADynamicSessionsCodeExecutor", - DeprecationWarning, - stacklevel=2, - ) - super().__init__(*args, **kwargs) - - -__all__ = [ - "AzureContainerCodeExecutor", -] diff --git a/python/packages/autogen-ext/src/autogen_ext/code_executor/docker_executor/__init__.py b/python/packages/autogen-ext/src/autogen_ext/code_executor/docker_executor/__init__.py deleted file mode 100644 index 66719114300d..000000000000 --- a/python/packages/autogen-ext/src/autogen_ext/code_executor/docker_executor/__init__.py +++ /dev/null @@ -1,11 +0,0 @@ -import warnings - -from ...code_executors import DockerCommandLineCodeExecutor - -warnings.warn( - "DockerCommandLineCodeExecutor moved to autogen_ext.code_executors.DockerCommandLineCodeExecutor", - DeprecationWarning, - stacklevel=2, -) - -__all__ = ["DockerCommandLineCodeExecutor"] diff --git a/python/packages/autogen-ext/src/autogen_ext/models/__init__.py b/python/packages/autogen-ext/src/autogen_ext/models/__init__.py index d39c1d9bf247..80533f80575e 100644 --- a/python/packages/autogen-ext/src/autogen_ext/models/__init__.py +++ b/python/packages/autogen-ext/src/autogen_ext/models/__init__.py @@ -2,6 +2,13 @@ AzureOpenAIChatCompletionClient, OpenAIChatCompletionClient, ) +from ._openai.config import AzureOpenAIClientConfiguration, OpenAIClientConfiguration from ._reply_chat_completion_client import ReplayChatCompletionClient -__all__ = ["AzureOpenAIChatCompletionClient", "OpenAIChatCompletionClient", "ReplayChatCompletionClient"] +__all__ = [ + "AzureOpenAIClientConfiguration", + "AzureOpenAIChatCompletionClient", + "OpenAIClientConfiguration", + "OpenAIChatCompletionClient", + "ReplayChatCompletionClient", +] diff --git a/python/packages/autogen-ext/src/autogen_ext/models/_openai/config/__init__.py b/python/packages/autogen-ext/src/autogen_ext/models/_openai/config/__init__.py index b6729a70d11e..53abfcc58796 100644 --- a/python/packages/autogen-ext/src/autogen_ext/models/_openai/config/__init__.py +++ b/python/packages/autogen-ext/src/autogen_ext/models/_openai/config/__init__.py @@ -49,3 +49,6 @@ class AzureOpenAIClientConfiguration(BaseOpenAIClientConfiguration, total=False) azure_ad_token_provider: AsyncAzureADTokenProvider # Must be provided model_capabilities: Required[ModelCapabilities] + + +__all__ = ["AzureOpenAIClientConfiguration", "OpenAIClientConfiguration"] diff --git a/python/packages/autogen-ext/src/autogen_ext/tools/langchain/__init__.py b/python/packages/autogen-ext/src/autogen_ext/tools/langchain/__init__.py deleted file mode 100644 index 4d401fc7ef1f..000000000000 --- a/python/packages/autogen-ext/src/autogen_ext/tools/langchain/__init__.py +++ /dev/null @@ -1,7 +0,0 @@ -import warnings - -from ...tools import LangChainToolAdapter - -warnings.warn("LangChainToolAdapter moved to autogen_ext.tools.LangChainToolAdapter", DeprecationWarning, stacklevel=2) - -__all__ = ["LangChainToolAdapter"] diff --git a/python/packages/autogen-ext/tests/models/test_openai_model_client.py b/python/packages/autogen-ext/tests/models/test_openai_model_client.py index a51e33c0234a..cee3be5835b7 100644 --- a/python/packages/autogen-ext/tests/models/test_openai_model_client.py +++ b/python/packages/autogen-ext/tests/models/test_openai_model_client.py @@ -1,5 +1,5 @@ import asyncio -from typing import Any, AsyncGenerator, List, Tuple +from typing import Annotated, Any, AsyncGenerator, List, Tuple from unittest.mock import MagicMock import pytest @@ -15,17 +15,25 @@ SystemMessage, UserMessage, ) -from autogen_core.components.tools import FunctionTool +from autogen_core.components.tools import BaseTool, FunctionTool from autogen_ext.models import AzureOpenAIChatCompletionClient, OpenAIChatCompletionClient from autogen_ext.models._openai._model_info import resolve_model -from autogen_ext.models._openai._openai_client import calculate_vision_tokens +from autogen_ext.models._openai._openai_client import calculate_vision_tokens, convert_tools from openai.resources.chat.completions import AsyncCompletions from openai.types.chat.chat_completion import ChatCompletion, Choice from openai.types.chat.chat_completion_chunk import ChatCompletionChunk, ChoiceDelta from openai.types.chat.chat_completion_chunk import Choice as ChunkChoice from openai.types.chat.chat_completion_message import ChatCompletionMessage from openai.types.completion_usage import CompletionUsage -from pydantic import BaseModel +from pydantic import BaseModel, Field + + +class MyResult(BaseModel): + result: str = Field(description="The other description.") + + +class MyArgs(BaseModel): + query: str = Field(description="The description.") class MockChunkDefinition(BaseModel): @@ -302,3 +310,38 @@ def test_openai_count_image_tokens(mock_size: Tuple[int, int], expected_num_toke # Directly call calculate_vision_tokens and check the result calculated_tokens = calculate_vision_tokens(mock_image, detail="auto") assert calculated_tokens == expected_num_tokens + + +def test_convert_tools_accepts_both_func_tool_and_schema() -> None: + def my_function(arg: str, other: Annotated[int, "int arg"], nonrequired: int = 5) -> MyResult: + return MyResult(result="test") + + tool = FunctionTool(my_function, description="Function tool.") + schema = tool.schema + + converted_tool_schema = convert_tools([tool, schema]) + + assert len(converted_tool_schema) == 2 + assert converted_tool_schema[0] == converted_tool_schema[1] + + +def test_convert_tools_accepts_both_tool_and_schema() -> None: + class MyTool(BaseTool[MyArgs, MyResult]): + def __init__(self) -> None: + super().__init__( + args_type=MyArgs, + return_type=MyResult, + name="TestTool", + description="Description of test tool.", + ) + + async def run(self, args: MyArgs, cancellation_token: CancellationToken) -> MyResult: + return MyResult(result="value") + + tool = MyTool() + schema = tool.schema + + converted_tool_schema = convert_tools([tool, schema]) + + assert len(converted_tool_schema) == 2 + assert converted_tool_schema[0] == converted_tool_schema[1] From 1e0b254d0aa011712328ac7f40b7788e37e63510 Mon Sep 17 00:00:00 2001 From: gagb Date: Fri, 22 Nov 2024 14:06:04 -0500 Subject: [PATCH 07/10] Update README.md with link to clarifications statement (#4318) --- README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/README.md b/README.md index 71509427170e..b5bda7de8e0e 100644 --- a/README.md +++ b/README.md @@ -10,7 +10,7 @@ # AutoGen > [!IMPORTANT] -> +> - (11/14/24) ⚠️ In response to a number of asks to clarify and distinguish between official AutoGen and its forks that created confusion, we issued a [clarification statement](https://github.com/microsoft/autogen/discussions/4217). > - (10/13/24) Interested in the standard AutoGen as a prior user? Find it at the actively-maintained *AutoGen* [0.2 branch](https://github.com/microsoft/autogen/tree/0.2) and `autogen-agentchat~=0.2` PyPi package. > - (10/02/24) [AutoGen 0.4](https://microsoft.github.io/autogen/dev) is a from-the-ground-up rewrite of AutoGen. Learn more about the history, goals and future at [this blog post](https://microsoft.github.io/autogen/blog). We’re excited to work with the community to gather feedback, refine, and improve the project before we officially release 0.4. This is a big change, so AutoGen 0.2 is still available, maintained, and developed in the [0.2 branch](https://github.com/microsoft/autogen/tree/0.2). From 8f4d8c89c3cceaf18540f1b4904ada13bc120393 Mon Sep 17 00:00:00 2001 From: Xiaoyun Zhang Date: Fri, 22 Nov 2024 13:51:08 -0800 Subject: [PATCH 08/10] .NET add roleplay tool call orchestrator in AutoGen.OpenAI (#4323) * add roleplay tool call orchestrator * add chinese business workflow test * update --- .../src/AutoGen.OpenAI/AutoGen.OpenAI.csproj | 1 + .../RolePlayToolCallOrchestrator.cs | 133 +++++++++ .../RolePlayToolCallOrchestratorTests.cs | 269 ++++++++++++++++++ 3 files changed, 403 insertions(+) create mode 100644 dotnet/src/AutoGen.OpenAI/Orchestrator/RolePlayToolCallOrchestrator.cs create mode 100644 dotnet/test/AutoGen.OpenAI.Tests/RolePlayToolCallOrchestratorTests.cs diff --git a/dotnet/src/AutoGen.OpenAI/AutoGen.OpenAI.csproj b/dotnet/src/AutoGen.OpenAI/AutoGen.OpenAI.csproj index 7f00b63be86c..70c0f2b0d1ce 100644 --- a/dotnet/src/AutoGen.OpenAI/AutoGen.OpenAI.csproj +++ b/dotnet/src/AutoGen.OpenAI/AutoGen.OpenAI.csproj @@ -18,6 +18,7 @@ + diff --git a/dotnet/src/AutoGen.OpenAI/Orchestrator/RolePlayToolCallOrchestrator.cs b/dotnet/src/AutoGen.OpenAI/Orchestrator/RolePlayToolCallOrchestrator.cs new file mode 100644 index 000000000000..f088e1748e66 --- /dev/null +++ b/dotnet/src/AutoGen.OpenAI/Orchestrator/RolePlayToolCallOrchestrator.cs @@ -0,0 +1,133 @@ +// Copyright (c) Microsoft Corporation. All rights reserved. +// RolePlayToolCallOrchestrator.cs + +using System; +using System.Collections.Generic; +using System.Linq; +using System.Text; +using System.Threading; +using System.Threading.Tasks; +using AutoGen.OpenAI.Extension; +using OpenAI.Chat; + +namespace AutoGen.OpenAI.Orchestrator; + +/// +/// Orchestrating group chat using role play tool call +/// +public partial class RolePlayToolCallOrchestrator : IOrchestrator +{ + public readonly ChatClient chatClient; + private readonly Graph? workflow; + + public RolePlayToolCallOrchestrator(ChatClient chatClient, Graph? workflow = null) + { + this.chatClient = chatClient; + this.workflow = workflow; + } + + public async Task GetNextSpeakerAsync( + OrchestrationContext context, + CancellationToken cancellationToken = default) + { + var candidates = context.Candidates.ToList(); + + if (candidates.Count == 0) + { + return null; + } + + if (candidates.Count == 1) + { + return candidates.First(); + } + + // if there's a workflow + // and the next available agent from the workflow is in the group chat + // then return the next agent from the workflow + if (this.workflow != null) + { + var lastMessage = context.ChatHistory.LastOrDefault(); + if (lastMessage == null) + { + return null; + } + var currentSpeaker = candidates.First(candidates => candidates.Name == lastMessage.From); + var nextAgents = await this.workflow.TransitToNextAvailableAgentsAsync(currentSpeaker, context.ChatHistory, cancellationToken); + nextAgents = nextAgents.Where(nextAgent => candidates.Any(candidate => candidate.Name == nextAgent.Name)); + candidates = nextAgents.ToList(); + if (!candidates.Any()) + { + return null; + } + + if (candidates is { Count: 1 }) + { + return candidates.First(); + } + } + + // In this case, since there are more than one available agents from the workflow for the next speaker + // We need to invoke LLM to select the next speaker via select next speaker function + + var chatHistoryStringBuilder = new StringBuilder(); + foreach (var message in context.ChatHistory) + { + var chatHistoryPrompt = $"{message.From}: {message.GetContent()}"; + + chatHistoryStringBuilder.AppendLine(chatHistoryPrompt); + } + + var chatHistory = chatHistoryStringBuilder.ToString(); + + var prompt = $""" + # Task: Select the next speaker + + You are in a role-play game. Carefully read the conversation history and select the next speaker from the available roles. + + # Conversation + {chatHistory} + + # Available roles + - {string.Join(",", candidates.Select(candidate => candidate.Name))} + + Select the next speaker from the available roles and provide a reason for your selection. + """; + + // enforce the next speaker to be selected by the LLM + var option = new ChatCompletionOptions + { + ToolChoice = ChatToolChoice.CreateFunctionChoice(this.SelectNextSpeakerFunctionContract.Name), + }; + + option.Tools.Add(this.SelectNextSpeakerFunctionContract.ToChatTool()); + var toolCallMiddleware = new FunctionCallMiddleware( + functions: [this.SelectNextSpeakerFunctionContract], + functionMap: new Dictionary>> + { + [this.SelectNextSpeakerFunctionContract.Name] = this.SelectNextSpeakerWrapper, + }); + + var selectAgent = new OpenAIChatAgent( + chatClient, + "admin", + option) + .RegisterMessageConnector() + .RegisterMiddleware(toolCallMiddleware); + + var reply = await selectAgent.SendAsync(prompt); + + var nextSpeaker = candidates.FirstOrDefault(candidate => candidate.Name == reply.GetContent()); + + return nextSpeaker; + } + + /// + /// Select the next speaker by name and reason + /// + [Function] + public async Task SelectNextSpeaker(string name, string reason) + { + return name; + } +} diff --git a/dotnet/test/AutoGen.OpenAI.Tests/RolePlayToolCallOrchestratorTests.cs b/dotnet/test/AutoGen.OpenAI.Tests/RolePlayToolCallOrchestratorTests.cs new file mode 100644 index 000000000000..807bf41e9479 --- /dev/null +++ b/dotnet/test/AutoGen.OpenAI.Tests/RolePlayToolCallOrchestratorTests.cs @@ -0,0 +1,269 @@ +// Copyright (c) Microsoft Corporation. All rights reserved. +// RolePlayToolCallOrchestratorTests.cs + +using System; +using System.Collections.Generic; +using System.Threading.Tasks; +using AutoGen.OpenAI.Orchestrator; +using AutoGen.Tests; +using Azure.AI.OpenAI; +using FluentAssertions; +using Moq; +using OpenAI; +using OpenAI.Chat; +using Xunit; + +namespace AutoGen.OpenAI.Tests; + +public class RolePlayToolCallOrchestratorTests +{ + [Fact] + public async Task ItReturnNullWhenNoCandidateIsAvailableAsync() + { + var chatClient = Mock.Of(); + var orchestrator = new RolePlayToolCallOrchestrator(chatClient); + var context = new OrchestrationContext + { + Candidates = [], + ChatHistory = [], + }; + + var speaker = await orchestrator.GetNextSpeakerAsync(context); + speaker.Should().BeNull(); + } + + [Fact] + public async Task ItReturnCandidateWhenOnlyOneCandidateIsAvailableAsync() + { + var chatClient = Mock.Of(); + var alice = new EchoAgent("Alice"); + var orchestrator = new RolePlayToolCallOrchestrator(chatClient); + var context = new OrchestrationContext + { + Candidates = [alice], + ChatHistory = [], + }; + + var speaker = await orchestrator.GetNextSpeakerAsync(context); + speaker.Should().Be(alice); + } + + [Fact] + public async Task ItSelectNextSpeakerFromWorkflowIfProvided() + { + var workflow = new Graph(); + var alice = new EchoAgent("Alice"); + var bob = new EchoAgent("Bob"); + var charlie = new EchoAgent("Charlie"); + workflow.AddTransition(Transition.Create(alice, bob)); + workflow.AddTransition(Transition.Create(bob, charlie)); + workflow.AddTransition(Transition.Create(charlie, alice)); + + var client = Mock.Of(); + var orchestrator = new RolePlayToolCallOrchestrator(client, workflow); + var context = new OrchestrationContext + { + Candidates = [alice, bob, charlie], + ChatHistory = + [ + new TextMessage(Role.User, "Hello, Bob", from: "Alice"), + ], + }; + + var speaker = await orchestrator.GetNextSpeakerAsync(context); + speaker.Should().Be(bob); + } + + [Fact] + public async Task ItReturnNullIfNoAvailableAgentFromWorkflowAsync() + { + var workflow = new Graph(); + var alice = new EchoAgent("Alice"); + var bob = new EchoAgent("Bob"); + workflow.AddTransition(Transition.Create(alice, bob)); + + var client = Mock.Of(); + var orchestrator = new RolePlayToolCallOrchestrator(client, workflow); + var context = new OrchestrationContext + { + Candidates = [alice, bob], + ChatHistory = + [ + new TextMessage(Role.User, "Hello, Alice", from: "Bob"), + ], + }; + + var speaker = await orchestrator.GetNextSpeakerAsync(context); + speaker.Should().BeNull(); + } + + [ApiKeyFact("AZURE_OPENAI_API_KEY", "AZURE_OPENAI_ENDPOINT", "AZURE_OPENAI_DEPLOY_NAME")] + public async Task GPT_3_5_CoderReviewerRunnerTestAsync() + { + var endpoint = Environment.GetEnvironmentVariable("AZURE_OPENAI_ENDPOINT") ?? throw new Exception("Please set AZURE_OPENAI_ENDPOINT environment variable."); + var key = Environment.GetEnvironmentVariable("AZURE_OPENAI_API_KEY") ?? throw new Exception("Please set AZURE_OPENAI_API_KEY environment variable."); + var deployName = Environment.GetEnvironmentVariable("AZURE_OPENAI_DEPLOY_NAME") ?? throw new Exception("Please set AZURE_OPENAI_DEPLOY_NAME environment variable."); + var openaiClient = new AzureOpenAIClient(new Uri(endpoint), new System.ClientModel.ApiKeyCredential(key)); + var chatClient = openaiClient.GetChatClient(deployName); + + await BusinessWorkflowTest(chatClient); + await CoderReviewerRunnerTestAsync(chatClient); + } + + [ApiKeyFact("OPENAI_API_KEY")] + public async Task GPT_4o_CoderReviewerRunnerTestAsync() + { + var apiKey = Environment.GetEnvironmentVariable("OPENAI_API_KEY") ?? throw new InvalidOperationException("OPENAI_API_KEY is not set"); + var model = "gpt-4o"; + var openaiClient = new OpenAIClient(apiKey); + var chatClient = openaiClient.GetChatClient(model); + + await BusinessWorkflowTest(chatClient); + await CoderReviewerRunnerTestAsync(chatClient); + } + + [ApiKeyFact("OPENAI_API_KEY")] + public async Task GPT_4o_mini_CoderReviewerRunnerTestAsync() + { + var apiKey = Environment.GetEnvironmentVariable("OPENAI_API_KEY") ?? throw new InvalidOperationException("OPENAI_API_KEY is not set"); + var model = "gpt-4o-mini"; + var openaiClient = new OpenAIClient(apiKey); + var chatClient = openaiClient.GetChatClient(model); + + await BusinessWorkflowTest(chatClient); + await CoderReviewerRunnerTestAsync(chatClient); + } + + /// + /// This test is to mimic the conversation among coder, reviewer and runner. + /// The coder will write the code, the reviewer will review the code, and the runner will run the code. + /// + /// + /// + private async Task CoderReviewerRunnerTestAsync(ChatClient client) + { + var coder = new EchoAgent("Coder"); + var reviewer = new EchoAgent("Reviewer"); + var runner = new EchoAgent("Runner"); + var user = new EchoAgent("User"); + var initializeMessage = new List + { + new TextMessage(Role.User, "Hello, I am user, I will provide the coding task, please write the code first, then review and run it", from: "User"), + new TextMessage(Role.User, "Hello, I am coder, I will write the code", from: "Coder"), + new TextMessage(Role.User, "Hello, I am reviewer, I will review the code", from: "Reviewer"), + new TextMessage(Role.User, "Hello, I am runner, I will run the code", from: "Runner"), + new TextMessage(Role.User, "how to print 'hello world' using C#", from: user.Name), + }; + + var chatHistory = new List() + { + new TextMessage(Role.User, """ + ```csharp + Console.WriteLine("Hello World"); + ``` + """, from: coder.Name), + new TextMessage(Role.User, "The code looks good", from: reviewer.Name), + new TextMessage(Role.User, "The code runs successfully, the output is 'Hello World'", from: runner.Name), + }; + + var orchestrator = new RolePlayToolCallOrchestrator(client); + foreach (var message in chatHistory) + { + var context = new OrchestrationContext + { + Candidates = [coder, reviewer, runner, user], + ChatHistory = initializeMessage, + }; + + var speaker = await orchestrator.GetNextSpeakerAsync(context); + speaker!.Name.Should().Be(message.From); + initializeMessage.Add(message); + } + + // the last next speaker should be the user + var lastSpeaker = await orchestrator.GetNextSpeakerAsync(new OrchestrationContext + { + Candidates = [coder, reviewer, runner, user], + ChatHistory = initializeMessage, + }); + + lastSpeaker!.Name.Should().Be(user.Name); + } + + // test if the tool call orchestrator still run business workflow when the conversation is not in English + private async Task BusinessWorkflowTest(ChatClient client) + { + var ceo = new EchoAgent("乙方首席执行官"); + var pm = new EchoAgent("乙方项目经理"); + var dev = new EchoAgent("乙方开发人员"); + var user = new EchoAgent("甲方"); + var initializeMessage = new List + { + new TextMessage(Role.User, "你好，我是你们的甲方", from: user.Name), + new TextMessage(Role.User, "你好，我是乙方首席执行官，我将负责对接甲方和给项目经理及开发人员分配任务", from: ceo.Name), + new TextMessage(Role.User, "你好，我是乙方项目经理，我将负责项目的进度和质量", from: pm.Name), + new TextMessage(Role.User, "你好，我是乙方开发人员 我将负责项目的具体开发", from: dev.Name), + new TextMessage(Role.User, "开发一个淘宝，预算1W", from: user.Name), + }; + + var workflow = new Graph(); + workflow.AddTransition(Transition.Create(ceo, pm)); + workflow.AddTransition(Transition.Create(ceo, dev)); + workflow.AddTransition(Transition.Create(pm, ceo)); + workflow.AddTransition(Transition.Create(dev, ceo)); + workflow.AddTransition(Transition.Create(user, ceo)); + workflow.AddTransition(Transition.Create(ceo, user)); + + var chatHistory = new List() + { + new TextMessage(Role.User, """ + 项目经理，如何使用1W预算开发一个淘宝 + """, from: ceo.Name), + new TextMessage(Role.User, """ + 对于1万预算开发淘宝类网站,以下是关键建议: + 技术选择: + - 使用开源电商系统节省成本, 选择便宜但稳定的云服务器和域名,预算2000元/年 + - 核心功能优先 + - 人员安排: + - 找1位全栈开发,负责系统搭建(6000元) + - 1位兼职UI设计(2000元) + - 进度规划: + - 基础功能1个月完成,后续根据运营情况逐步优化。 + """, from: pm.Name), + new TextMessage(Role.User, "好的，开发人员，请根据项目经理的规划开始开发", from: ceo.Name), + new TextMessage(Role.User, """ + 好的，已开发完毕 + ```html + + ``` + """, from: dev.Name), + new TextMessage(Role.User, "好的，项目已完成，甲方请付款", from: ceo.Name), + }; + + var orchestrator = new RolePlayToolCallOrchestrator(client, workflow); + + foreach (var message in chatHistory) + { + var context = new OrchestrationContext + { + Candidates = [ceo, pm, dev, user], + ChatHistory = initializeMessage, + }; + + var speaker = await orchestrator.GetNextSpeakerAsync(context); + speaker!.Name.Should().Be(message.From); + initializeMessage.Add(message); + } + + // the last next speaker should be the user + var lastSpeaker = await orchestrator.GetNextSpeakerAsync(new OrchestrationContext + { + Candidates = [ceo, pm, dev, user], + ChatHistory = initializeMessage, + }); + + lastSpeaker!.Name.Should().Be(user.Name); + } +} From 0b5eaf1240e05f0ef05595c92c0d33a9850d0dc4 Mon Sep 17 00:00:00 2001 From: Thai Nguyen Date: Sat, 23 Nov 2024 22:07:21 +0700 Subject: [PATCH 09/10] Agent name termination (#4123) --- .../src/autogen_agentchat/task/__init__.py | 2 ++ .../autogen_agentchat/task/_terminations.py | 35 ++++++++++++++++++- .../tests/test_termination_condition.py | 25 +++++++++++++ 3 files changed, 61 insertions(+), 1 deletion(-) diff --git a/python/packages/autogen-agentchat/src/autogen_agentchat/task/__init__.py b/python/packages/autogen-agentchat/src/autogen_agentchat/task/__init__.py index dd7b6265ad44..e1e6766338d3 100644 --- a/python/packages/autogen-agentchat/src/autogen_agentchat/task/__init__.py +++ b/python/packages/autogen-agentchat/src/autogen_agentchat/task/__init__.py @@ -7,6 +7,7 @@ TextMentionTermination, TimeoutTermination, TokenUsageTermination, + SourceMatchTermination, ) __all__ = [ @@ -17,5 +18,6 @@ "HandoffTermination", "TimeoutTermination", "ExternalTermination", + "SourceMatchTermination", "Console", ] diff --git a/python/packages/autogen-agentchat/src/autogen_agentchat/task/_terminations.py b/python/packages/autogen-agentchat/src/autogen_agentchat/task/_terminations.py index f8d79cef2850..81cb5cca7d6c 100644 --- a/python/packages/autogen-agentchat/src/autogen_agentchat/task/_terminations.py +++ b/python/packages/autogen-agentchat/src/autogen_agentchat/task/_terminations.py @@ -1,5 +1,5 @@ import time -from typing import Sequence +from typing import Sequence, List from ..base import TerminatedException, TerminationCondition from ..messages import AgentMessage, HandoffMessage, MultiModalMessage, StopMessage, TextMessage @@ -251,3 +251,36 @@ async def __call__(self, messages: Sequence[AgentMessage]) -> StopMessage | None async def reset(self) -> None: self._terminated = False self._setted = False + + +class SourceMatchTermination(TerminationCondition): + """Terminate the conversation after a specific source responds. + + Args: + sources (List[str]): List of source names to terminate the conversation. + + Raises: + TerminatedException: If the termination condition has already been reached. + """ + + def __init__(self, sources: List[str]) -> None: + self._sources = sources + self._terminated = False + + @property + def terminated(self) -> bool: + return self._terminated + + async def __call__(self, messages: Sequence[AgentMessage]) -> StopMessage | None: + if self._terminated: + raise TerminatedException("Termination condition has already been reached") + if not messages: + return None + for message in messages: + if message.source in self._sources: + self._terminated = True + return StopMessage(content=f"'{message.source}' answered", source="SourceMatchTermination") + return None + + async def reset(self) -> None: + self._terminated = False diff --git a/python/packages/autogen-agentchat/tests/test_termination_condition.py b/python/packages/autogen-agentchat/tests/test_termination_condition.py index c09e0e1c14ac..f4aa5d2a7203 100644 --- a/python/packages/autogen-agentchat/tests/test_termination_condition.py +++ b/python/packages/autogen-agentchat/tests/test_termination_condition.py @@ -1,6 +1,7 @@ import asyncio import pytest +from autogen_agentchat.base import TerminatedException from autogen_agentchat.messages import HandoffMessage, StopMessage, TextMessage from autogen_agentchat.task import ( ExternalTermination, @@ -10,6 +11,7 @@ TextMentionTermination, TimeoutTermination, TokenUsageTermination, + SourceMatchTermination, ) from autogen_core.components.models import RequestUsage @@ -242,3 +244,26 @@ async def test_external_termination() -> None: await termination.reset() assert await termination([]) is None + + +@pytest.mark.asyncio +async def test_source_match_termination() -> None: + termination = SourceMatchTermination(sources=["Assistant"]) + assert await termination([]) is None + + continue_messages = [TextMessage(content="Hello", source="agent"), TextMessage(content="Hello", source="user")] + assert await termination(continue_messages) is None + + terminate_messages = [ + TextMessage(content="Hello", source="agent"), + TextMessage(content="Hello", source="Assistant"), + TextMessage(content="Hello", source="user"), + ] + result = await termination(terminate_messages) + assert isinstance(result, StopMessage) + assert termination.terminated + + with pytest.raises(TerminatedException): + await termination([]) + await termination.reset() + assert not termination.terminated From caeab68f4b577212529781f3bad7626f9d55d204 Mon Sep 17 00:00:00 2001 From: Pramod Goyal <81946962+goyalpramod@users.noreply.github.com> Date: Sun, 24 Nov 2024 04:56:37 +0530 Subject: [PATCH 10/10] task: added warning when none is called in intervention handler (#4149) * task: added warning when none is called in intervention handler * add leading underscore to indicate private to _warn_if_none method in intervention.py * address comment of returning Any for result in intervention.py * Update intervention.py to remove redundant name change * Format and lint --------- Co-authored-by: Jack Gerrits Co-authored-by: Jack Gerrits --- .../src/autogen_core/base/intervention.py | 21 +++++++++++++++++++ 1 file changed, 21 insertions(+) diff --git a/python/packages/autogen-core/src/autogen_core/base/intervention.py b/python/packages/autogen-core/src/autogen_core/base/intervention.py index c9600ac9e13c..4b06fa19f94f 100644 --- a/python/packages/autogen-core/src/autogen_core/base/intervention.py +++ b/python/packages/autogen-core/src/autogen_core/base/intervention.py @@ -1,3 +1,4 @@ +import warnings from typing import Any, Awaitable, Callable, Protocol, final from autogen_core.base import AgentId @@ -14,6 +15,23 @@ class DropMessage: ... +def _warn_if_none(value: Any, handler_name: str) -> None: + """ + Utility function to check if the intervention handler returned None and issue a warning. + + Args: + value: The return value to check + handler_name: Name of the intervention handler method for the warning message + """ + if value is None: + warnings.warn( + f"Intervention handler {handler_name} returned None. This might be unintentional. " + "Consider returning the original message or DropMessage explicitly.", + RuntimeWarning, + stacklevel=2, + ) + + InterventionFunction = Callable[[Any], Any | Awaitable[type[DropMessage]]] @@ -27,10 +45,13 @@ async def on_response( class DefaultInterventionHandler(InterventionHandler): async def on_send(self, message: Any, *, sender: AgentId | None, recipient: AgentId) -> Any | type[DropMessage]: + _warn_if_none(message, "on_send") return message async def on_publish(self, message: Any, *, sender: AgentId | None) -> Any | type[DropMessage]: + _warn_if_none(message, "on_publish") return message async def on_response(self, message: Any, *, sender: AgentId, recipient: AgentId | None) -> Any | type[DropMessage]: + _warn_if_none(message, "on_response") return message