langroid 0.16.0__tar.gz → 0.16.2__tar.gz

{langroid-0.16.0 → langroid-0.16.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: langroid
-Version: 0.16.0
+Version: 0.16.2
 Summary: Harness LLMs with Multi-Agent Programming
 License: MIT
 Author: Prasad Chalasani
@@ -242,6 +242,11 @@ teacher_task.run()
 <details>
 <summary> <b>Click to expand</b></summary>
 
+- **Sep 2024:**
+  - **[0.16.0](https://github.com/langroid/langroid/releases/tag/0.16.0)**  Support for OpenAI `o1-mini` and `o1-preview` models.
+  - **[0.15.0](https://github.com/langroid/langroid/releases/tag/0.15.0)** Cerebras API support -- run llama-3.1 models hosted on Cerebras Cloud (very fast inference).
+  - **[0.14.0](https://github.com/langroid/langroid/releases/tag/0.14.0)** `DocChatAgent` uses Reciprocal Rank Fusion (RRF) to rank chunks retrieved by different methods.
+  - **[0.12.0](https://github.com/langroid/langroid/releases/tag/0.12.0)** `run_batch_task` new option -- `stop_on_first_result` - allows termination of batch as soon as any task returns a result.  
 - **Aug 2024:**
   - **[0.11.0](https://github.com/langroid/langroid/releases/tag/0.11.0)** Polymorphic `Task.run(), Task.run_async`.
   - **[0.10.0](https://github.com/langroid/langroid/releases/tag/0.10.0)** Allow tool handlers to return arbitrary result type, including other tools.

{langroid-0.16.0 → langroid-0.16.2}/README.md

@@ -134,6 +134,11 @@ teacher_task.run()
 <details>
 <summary> <b>Click to expand</b></summary>
 
+- **Sep 2024:**
+  - **[0.16.0](https://github.com/langroid/langroid/releases/tag/0.16.0)**  Support for OpenAI `o1-mini` and `o1-preview` models.
+  - **[0.15.0](https://github.com/langroid/langroid/releases/tag/0.15.0)** Cerebras API support -- run llama-3.1 models hosted on Cerebras Cloud (very fast inference).
+  - **[0.14.0](https://github.com/langroid/langroid/releases/tag/0.14.0)** `DocChatAgent` uses Reciprocal Rank Fusion (RRF) to rank chunks retrieved by different methods.
+  - **[0.12.0](https://github.com/langroid/langroid/releases/tag/0.12.0)** `run_batch_task` new option -- `stop_on_first_result` - allows termination of batch as soon as any task returns a result.  
 - **Aug 2024:**
   - **[0.11.0](https://github.com/langroid/langroid/releases/tag/0.11.0)** Polymorphic `Task.run(), Task.run_async`.
   - **[0.10.0](https://github.com/langroid/langroid/releases/tag/0.10.0)** Allow tool handlers to return arbitrary result type, including other tools.

{langroid-0.16.0 → langroid-0.16.2}/langroid/agent/chat_agent.py

@@ -226,6 +226,7 @@ class ChatAgent(Agent):
             self.llm is not None
             and isinstance(self.llm, OpenAIGPT)
             and self.llm.is_openai_chat_model()
+            and self.llm.supports_functions_or_tools()
         )
 
     def set_system_message(self, msg: str) -> None:

{langroid-0.16.0 → langroid-0.16.2}/langroid/agent/task.py

@@ -98,6 +98,12 @@ class TaskConfig(BaseModel):
             (Search for "SEND_TO" in the examples/ dir to see how this is used.)
         allow_subtask_multi_oai_tools (bool): whether to allow multiple OpenAI
             tool-calls to be sent to a sub-task.
+        recognize_string_signals (bool): whether to recognize string-based signaling
+            like DONE, SEND_TO, PASS, etc. Default is True, but note that we don't need
+            to use string-based signaling, and it is recommended to use the
+            new Orchestration tools instead (see agent/tools/orchestration.py),
+            e.g. DoneTool, SendTool, etc.
+
     """
 
     inf_loop_cycle_len: int = 10
@@ -107,6 +113,7 @@ class TaskConfig(BaseModel):
     logs_dir: str = "logs"
     addressing_prefix: str = ""
     allow_subtask_multi_oai_tools: bool = True
+    recognize_string_signals: bool = True
 
 
 class Task:
@@ -1363,7 +1370,10 @@ class Task:
             else:
                 # e must be this agent's Entity (LLM, AGENT or USER)
                 return self.agent.response_template(e=e, tool_messages=[result])
-        # if result content starts with @name, set recipient to name
+        if not self.config.recognize_string_signals:
+            # ignore all string-based signaling/routing
+            return result
+        # parse various routing/addressing strings in result
         is_pass, recipient, content = parse_routing(
             result,
             addressing_prefix=self.config.addressing_prefix,
@@ -1490,7 +1500,7 @@ class Task:
 
         content = result_msg.content if result_msg else ""
         content_any = result_msg.content_any if result_msg else None
-        if DONE in content:
+        if DONE in content and self.config.recognize_string_signals:
             # assuming it is of the form "DONE: <content>"
             content = content.replace(DONE, "").strip()
         oai_tool_calls = result_msg.oai_tool_calls if result_msg else None
@@ -1556,12 +1566,14 @@ class Task:
         Returns:
             bool: True if msg is (equivalent to) empty or None, False otherwise
         """
+        # if ignoring string-based signaling, set pass_str to ""
+        pass_str = PASS if self.config.recognize_string_signals else ""
         return (
             msg is None
-            or (isinstance(msg, str) and msg.strip() in [PASS, ""])
+            or (isinstance(msg, str) and msg.strip() in [pass_str, ""])
             or (
                 isinstance(msg, ChatDocument)
-                and msg.content.strip() in [PASS, ""]
+                and msg.content.strip() in [pass_str, ""]
                 and msg.function_call is None
                 and msg.oai_tool_calls is None
                 and msg.oai_tool_id2result is None
@@ -1574,12 +1586,13 @@ class Task:
     ) -> bool:
         """Is the task done based on the response from the given responder?"""
 
+        allow_done_string = self.config.recognize_string_signals
         response_says_done = result is not None and (
-            (isinstance(result, str) and DONE in result)
+            (isinstance(result, str) and DONE in result and allow_done_string)
             or (
                 isinstance(result, ChatDocument)
                 and (
-                    DONE in result.content
+                    (DONE in result.content and allow_done_string)
                     or any(
                         isinstance(t, (DoneTool, AgentDoneTool, FinalResultTool))
                         for t in result.tool_messages
@@ -1697,10 +1710,14 @@ class Task:
         if self._is_kill():
             return (True, StatusCode.KILL)
         result = result or self.pending_message
+        allow_done_string = self.config.recognize_string_signals
         # An entity decided task is done, either via DoneTool,
         # or by explicitly saying DONE
         done_result = result is not None and (
-            DONE in (result.content if isinstance(result, str) else result.content)
+            (
+                DONE in (result.content if isinstance(result, str) else result.content)
+                and allow_done_string
+            )
             or any(
                 isinstance(t, (DoneTool, AgentDoneTool, FinalResultTool))
                 for t in result.tool_messages

{langroid-0.16.0 → langroid-0.16.2}/langroid/language_models/base.py

@@ -201,16 +201,24 @@ class LLMMessage(BaseModel):
     # link to corresponding chat document, for provenance/rewind purposes
     chat_document_id: str = ""
 
-    def api_dict(self) -> Dict[str, Any]:
+    def api_dict(self, has_system_role: bool = True) -> Dict[str, Any]:
         """
         Convert to dictionary for API request, keeping ONLY
         the fields that are expected in an API call!
         E.g., DROP the tool_id, since it is only for use in the Assistant API,
             not the completion API.
+
+        Args:
+            has_system_role: whether the message has a system role (if not,
+                set to "user" role)
         Returns:
             dict: dictionary representation of LLM message
         """
         d = self.dict()
+        # if there is a key k = "role" with value "system", change to "user"
+        # in case has_system_role is False
+        if not has_system_role and "role" in d and d["role"] == "system":
+            d["role"] = "user"
         # drop None values since API doesn't accept them
         dict_no_none = {k: v for k, v in d.items() if v is not None}
         if "name" in dict_no_none and dict_no_none["name"] == "":

{langroid-0.16.0 → langroid-0.16.2}/langroid/language_models/openai_gpt.py

@@ -592,6 +592,12 @@ class OpenAIGPT(LanguageModel):
         openai_chat_models = [e.value for e in OpenAIChatModel]
         return self.config.chat_model in openai_chat_models
 
+    def supports_functions_or_tools(self) -> bool:
+        return self.is_openai_chat_model() and self.config.chat_model not in [
+            OpenAIChatModel.O1_MINI,
+            OpenAIChatModel.O1_PREVIEW,
+        ]
+
     def is_openai_completion_model(self) -> bool:
         openai_completion_models = [e.value for e in OpenAICompletionModel]
         return self.config.completion_model in openai_completion_models
@@ -606,6 +612,17 @@ class OpenAIGPT(LanguageModel):
             case _:
                 return []
 
+    def rename_params(self) -> Dict[str, str]:
+        """
+        Map of param name -> new name for specific models.
+        Currently main troublemaker is o1* series.
+        """
+        match self.config.chat_model:
+            case OpenAIChatModel.O1_MINI | OpenAIChatModel.O1_PREVIEW:
+                return {"max_tokens": "max_completion_tokens"}
+            case _:
+                return {}
+
     def chat_context_length(self) -> int:
         """
         Context-length for chat-completion models/endpoints
@@ -1186,7 +1203,7 @@ class OpenAIGPT(LanguageModel):
             kwargs["prompt"] = prompt
         args = dict(
             **kwargs,
-            max_completion_tokens=max_tokens,  # for output/completion
+            max_tokens=max_tokens,  # for output/completion
             stream=self.get_stream(),
         )
         args = self._openai_api_call_params(args)
@@ -1261,7 +1278,7 @@ class OpenAIGPT(LanguageModel):
             kwargs["prompt"] = prompt
         cached, hashed_key, response = await completions_with_backoff(
             **kwargs,
-            max_completion_tokens=max_tokens,
+            max_tokens=max_tokens,
             stream=False,
         )
         if not isinstance(response, dict):
@@ -1452,12 +1469,6 @@ class OpenAIGPT(LanguageModel):
         else:
             llm_messages = messages
 
-        # for models that don't support system msg, change SYSTEM role to USER
-        if self.config.chat_model in NON_SYSTEM_MESSAGE_MODELS:
-            for i, m in enumerate(llm_messages):
-                if m.role == Role.SYSTEM:
-                    llm_messages[i].role = Role.USER
-
         # Azure uses different parameters. It uses ``engine`` instead of ``model``
         # and the value should be the deployment_name not ``self.config.chat_model``
         chat_model = self.config.chat_model
@@ -1467,8 +1478,14 @@ class OpenAIGPT(LanguageModel):
 
         args: Dict[str, Any] = dict(
             model=chat_model,
-            messages=[m.api_dict() for m in llm_messages],
-            max_completion_tokens=max_tokens,
+            messages=[
+                m.api_dict(
+                    has_system_role=self.config.chat_model
+                    not in NON_SYSTEM_MESSAGE_MODELS
+                )
+                for m in (llm_messages)
+            ],
+            max_tokens=max_tokens,
             stream=self.get_stream(),
         )
         args.update(self._openai_api_call_params(args))
@@ -1498,6 +1515,11 @@ class OpenAIGPT(LanguageModel):
             # some models e.g. o1-mini (as of sep 2024) don't support some params,
             # like temperature and stream, so we need to remove them.
             args.pop(p, None)
+
+        param_rename_map = self.rename_params()
+        for old_param, new_param in param_rename_map.items():
+            if old_param in args:
+                args[new_param] = args.pop(old_param)
         return args
 
     def _process_chat_completion_response(

{langroid-0.16.0 → langroid-0.16.2}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "langroid"
-version = "0.16.0"
+version = "0.16.2"
 description = "Harness LLMs with Multi-Agent Programming"
 authors = ["Prasad Chalasani <pchalasani@gmail.com>"]
 readme = "README.md"