Merge branch 'master' into terminal-deps

Author: JINO-ROHIT

camel/agents/chat_agent.py

@@ -96,13 +96,12 @@ class DeepSeekConfig(BaseConfig):
     tool_choice: Optional[Union[dict[str, str], str]] = None
     logprobs: Optional[bool] = None
     top_logprobs: Optional[int] = None
+    stream_options: Optional[dict[str, bool]] = None
 
     def __init__(self, include_usage: bool = True, **kwargs):
+        if kwargs.get("stream") and "stream_options" not in kwargs:
+            kwargs["stream_options"] = {"include_usage": include_usage}
         super().__init__(**kwargs)
-        # Only set stream_options when stream is True
-        # Otherwise, it will raise error when calling the API
-        if self.stream:
-            self.stream_options = {"include_usage": include_usage}
 
 
 DEEPSEEK_API_PARAMS = {param for param in DeepSeekConfig.model_fields.keys()}

camel/configs/deepseek_config.py

@@ -71,8 +71,10 @@ class BaseMessage:
             images associated with the message. (default: :obj:`auto`)
         video_detail (Literal["auto", "low", "high"]): Detail level of the
             videos associated with the message. (default: :obj:`auto`)
-        parsed: Optional[Union[Type[BaseModel], dict]]: Optional object which
+        parsed (Optional[Union[Type[BaseModel], dict]]): Optional object which
             is parsed from the content. (default: :obj:`None`)
+        reasoning_content (Optional[str]): Optional reasoning trace associated
+            with the message. (default: :obj:`None`)
     """
 
     role_name: str
@@ -85,6 +87,7 @@ class BaseMessage:
     image_detail: Literal["auto", "low", "high"] = "auto"
     video_detail: Literal["auto", "low", "high"] = "auto"
     parsed: Optional[Union[BaseModel, dict]] = None
+    reasoning_content: Optional[str] = None
 
     @classmethod
     def make_user_message(
@@ -219,6 +222,12 @@ def create_new_instance(self, content: str) -> "BaseMessage":
             role_type=self.role_type,
             meta_dict=self.meta_dict,
             content=content,
+            video_bytes=self.video_bytes,
+            image_list=self.image_list,
+            image_detail=self.image_detail,
+            video_detail=self.video_detail,
+            parsed=self.parsed,
+            reasoning_content=self.reasoning_content,
         )
 
     def __add__(self, other: Any) -> Union["BaseMessage", Any]:

camel/messages/base.py

@@ -165,44 +165,6 @@ def _prepare_request(
 
         return request_config
 
-    def _post_handle_response(
-        self, response: ChatCompletion
-    ) -> ChatCompletion:
-        r"""Handle reasoning content with <think> tags at the beginning."""
-        if (
-            self.model_type in [ModelType.DEEPSEEK_REASONER]
-            and os.environ.get("GET_REASONING_CONTENT", "false").lower()
-            == "true"
-        ):
-            reasoning_content = response.choices[0].message.reasoning_content  # type: ignore[attr-defined]
-            combined_content = (  # type: ignore[operator]
-                f"<think>\n{reasoning_content}\n</think>\n"
-                if reasoning_content
-                else ""
-            ) + response.choices[0].message.content
-
-            response = ChatCompletion.construct(
-                id=response.id,
-                choices=[
-                    dict(
-                        index=response.choices[0].index,
-                        message={
-                            "role": response.choices[0].message.role,
-                            "content": combined_content,
-                            "tool_calls": None,
-                        },
-                        finish_reason=response.choices[0].finish_reason
-                        if response.choices[0].finish_reason
-                        else None,
-                    )
-                ],
-                created=response.created,
-                model=response.model,
-                object="chat.completion",
-                usage=response.usage,
-            )
-        return response
-
     @observe()
     def _run(
         self,
@@ -244,7 +206,7 @@ def _run(
             **request_config,
         )
 
-        return self._post_handle_response(response)
+        return response
 
     @observe()
     async def _arun(
@@ -286,4 +248,4 @@ async def _arun(
             **request_config,
         )
 
-        return self._post_handle_response(response)
+        return response

camel/models/deepseek_model.py

@@ -13,19 +13,11 @@
 # ========= Copyright 2023-2024 @ CAMEL-AI.org. All Rights Reserved. =========
 
 import os
-import time
-from typing import Any, Dict, List, Optional, Union
-
-from openai import AsyncStream, Stream
+from typing import Any, Dict, Optional, Union
 
 from camel.configs import ModelScopeConfig
-from camel.messages import OpenAIMessage
 from camel.models.openai_compatible_model import OpenAICompatibleModel
-from camel.types import (
-    ChatCompletion,
-    ChatCompletionChunk,
-    ModelType,
-)
+from camel.types import ModelType
 from camel.utils import (
     BaseTokenCounter,
     api_keys_required,
@@ -96,168 +88,3 @@ def __init__(
             max_retries=max_retries,
             **kwargs,
         )
-
-    def _post_handle_response(
-        self, response: Union[ChatCompletion, Stream[ChatCompletionChunk]]
-    ) -> ChatCompletion:
-        r"""Handle reasoning content with <think> tags at the beginning."""
-        if not isinstance(response, Stream):
-            # Handle non-streaming response (existing logic)
-            if self.model_config_dict.get("extra_body", {}).get(
-                "enable_thinking", False
-            ):
-                reasoning_content = response.choices[
-                    0
-                ].message.reasoning_content  # type: ignore[attr-defined]
-                combined_content = (
-                    f"<think>\n{reasoning_content}\n</think>\n"
-                    if reasoning_content
-                    else ""
-                )
-                response_content = response.choices[0].message.content or ""
-                combined_content += response_content
-
-                # Construct a new ChatCompletion with combined content
-                return ChatCompletion.construct(
-                    id=response.id,
-                    choices=[
-                        dict(
-                            finish_reason=response.choices[0].finish_reason,
-                            index=response.choices[0].index,
-                            logprobs=response.choices[0].logprobs,
-                            message=dict(
-                                role=response.choices[0].message.role,
-                                content=combined_content,
-                            ),
-                        )
-                    ],
-                    created=response.created,
-                    model=response.model,
-                    object="chat.completion",
-                    system_fingerprint=response.system_fingerprint,
-                    usage=response.usage,
-                )
-            else:
-                return response  # Return original if no thinking enabled
-
-        # Handle streaming response
-        accumulated_reasoning = ""
-        accumulated_content = ""
-        final_chunk = None
-        usage_data = None  # Initialize usage data
-        role = "assistant"  # Default role
-
-        for chunk in response:
-            final_chunk = chunk  # Keep track of the last chunk for metadata
-            if chunk.choices:
-                delta = chunk.choices[0].delta
-                if delta.role:
-                    role = delta.role  # Update role if provided
-                if (
-                    hasattr(delta, 'reasoning_content')
-                    and delta.reasoning_content
-                ):
-                    accumulated_reasoning += delta.reasoning_content
-                if delta.content:
-                    accumulated_content += delta.content
-
-            if hasattr(chunk, 'usage') and chunk.usage:
-                usage_data = chunk.usage
-
-        combined_content = (
-            f"<think>\n{accumulated_reasoning}\n</think>\n"
-            if accumulated_reasoning
-            else ""
-        ) + accumulated_content
-
-        # Construct the final ChatCompletion object from accumulated
-        # stream data
-        if final_chunk:
-            finish_reason = "stop"  # Default finish reason
-            logprobs = None
-            if final_chunk.choices:
-                finish_reason = (
-                    final_chunk.choices[0].finish_reason or finish_reason
-                )
-                if hasattr(final_chunk.choices[0], 'logprobs'):
-                    logprobs = final_chunk.choices[0].logprobs
-
-            return ChatCompletion.construct(
-                # Use data from the final chunk or defaults
-                id=final_chunk.id
-                if hasattr(final_chunk, 'id')
-                else "streamed-completion",
-                choices=[
-                    dict(
-                        finish_reason=finish_reason,
-                        index=0,
-                        logprobs=logprobs,
-                        message=dict(
-                            role=role,
-                            content=combined_content,
-                        ),
-                    )
-                ],
-                created=final_chunk.created
-                if hasattr(final_chunk, 'created')
-                else int(time.time()),
-                model=final_chunk.model
-                if hasattr(final_chunk, 'model')
-                else self.model_type,
-                object="chat.completion",
-                system_fingerprint=final_chunk.system_fingerprint
-                if hasattr(final_chunk, 'system_fingerprint')
-                else None,
-                usage=usage_data,
-            )
-        else:
-            # Handle cases where the stream was empty or invalid
-            return ChatCompletion.construct(
-                id="empty-stream",
-                choices=[
-                    dict(
-                        finish_reason="error",
-                        index=0,
-                        message=dict(role="assistant", content=""),
-                    )
-                ],
-                created=int(time.time()),
-                model=self.model_type,
-                object="chat.completion",
-                usage=usage_data,
-            )
-
-    def _request_chat_completion(
-        self,
-        messages: List[OpenAIMessage],
-        tools: Optional[List[Dict[str, Any]]] = None,
-    ) -> Union[ChatCompletion, Stream[ChatCompletionChunk]]:
-        request_config = self.model_config_dict.copy()
-
-        if tools:
-            request_config["tools"] = tools
-
-        return self._post_handle_response(
-            self._client.chat.completions.create(
-                messages=messages,
-                model=self.model_type,
-                **request_config,
-            )
-        )
-
-    async def _arequest_chat_completion(
-        self,
-        messages: List[OpenAIMessage],
-        tools: Optional[List[Dict[str, Any]]] = None,
-    ) -> Union[ChatCompletion, AsyncStream[ChatCompletionChunk]]:
-        request_config = self.model_config_dict.copy()
-
-        if tools:
-            request_config["tools"] = tools
-
-        response = await self._async_client.chat.completions.create(
-            messages=messages,
-            model=self.model_type,
-            **request_config,
-        )
-        return self._post_handle_response(response)