langroid 0.15.2__tar.gz → 0.16.1__tar.gz

{langroid-0.15.2 → langroid-0.16.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: langroid
-Version: 0.15.2
+Version: 0.16.1
 Summary: Harness LLMs with Multi-Agent Programming
 License: MIT
 Author: Prasad Chalasani
@@ -66,7 +66,7 @@ Requires-Dist: neo4j (>=5.14.1,<6.0.0) ; extra == "all" or extra == "neo4j"
 Requires-Dist: nest-asyncio (>=1.6.0,<2.0.0)
 Requires-Dist: nltk (>=3.8.1,<4.0.0)
 Requires-Dist: onnxruntime (>=1.16.1,<2.0.0)
-Requires-Dist: openai (>=1.14.0,<2.0.0)
+Requires-Dist: openai (>=1.45.0,<2.0.0)
 Requires-Dist: pandas (>=2.0.3,<3.0.0)
 Requires-Dist: pdf2image (>=1.17.0,<2.0.0) ; extra == "doc-chat" or extra == "all" or extra == "pdf-parsers"
 Requires-Dist: pdfplumber (>=0.10.2,<0.11.0) ; extra == "doc-chat" or extra == "all" or extra == "pdf-parsers"
@@ -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-0.15.2 → langroid-0.16.1}/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-0.15.2 → langroid-0.16.1}/langroid/language_models/openai_gpt.py

@@ -88,6 +88,8 @@ class OpenAIChatModel(str, Enum):
     GPT4_TURBO = "gpt-4-turbo"
     GPT4o = "gpt-4o-2024-08-06"
     GPT4o_MINI = "gpt-4o-mini"
+    O1_PREVIEW = "o1-preview"
+    O1_MINI = "o1-mini"
 
 
 class OpenAICompletionModel(str, Enum):
@@ -105,6 +107,8 @@ _context_length: Dict[str, int] = {
     OpenAIChatModel.GPT4_TURBO: 128_000,
     OpenAIChatModel.GPT4o: 128_000,
     OpenAIChatModel.GPT4o_MINI: 128_000,
+    OpenAIChatModel.O1_PREVIEW: 128_000,
+    OpenAIChatModel.O1_MINI: 128_000,
     OpenAICompletionModel.TEXT_DA_VINCI_003: 4096,
     AnthropicModel.CLAUDE_3_5_SONNET: 200_000,
     AnthropicModel.CLAUDE_3_OPUS: 200_000,
@@ -120,6 +124,8 @@ _cost_per_1k_tokens: Dict[str, Tuple[float, float]] = {
     OpenAIChatModel.GPT4_TURBO: (0.01, 0.03),  # 128K context
     OpenAIChatModel.GPT4o: (0.0025, 0.010),  # 128K context
     OpenAIChatModel.GPT4o_MINI: (0.00015, 0.0006),  # 128K context
+    OpenAIChatModel.O1_PREVIEW: (0.015, 0.060),  # 128K context
+    OpenAIChatModel.O1_MINI: (0.003, 0.012),  # 128K context
     AnthropicModel.CLAUDE_3_5_SONNET: (0.003, 0.015),
     AnthropicModel.CLAUDE_3_OPUS: (0.015, 0.075),
     AnthropicModel.CLAUDE_3_SONNET: (0.003, 0.015),
@@ -132,6 +138,8 @@ openAIChatModelPreferenceList = [
     OpenAIChatModel.GPT4_TURBO,
     OpenAIChatModel.GPT4,
     OpenAIChatModel.GPT4o_MINI,
+    OpenAIChatModel.O1_MINI,
+    OpenAIChatModel.O1_PREVIEW,
     OpenAIChatModel.GPT3_5_TURBO,
 ]
 
@@ -140,6 +148,15 @@ openAICompletionModelPreferenceList = [
     OpenAICompletionModel.TEXT_DA_VINCI_003,
 ]
 
+NON_STREAMING_MODELS = [
+    OpenAIChatModel.O1_MINI,
+    OpenAIChatModel.O1_PREVIEW,
+]
+
+NON_SYSTEM_MESSAGE_MODELS = [
+    OpenAIChatModel.O1_MINI,
+    OpenAIChatModel.O1_PREVIEW,
+]
 
 if "OPENAI_API_KEY" in os.environ:
     try:
@@ -579,6 +596,27 @@ class OpenAIGPT(LanguageModel):
         openai_completion_models = [e.value for e in OpenAICompletionModel]
         return self.config.completion_model in openai_completion_models
 
+    def unsupported_params(self) -> List[str]:
+        """
+        List of params that are not supported by the current model
+        """
+        match self.config.chat_model:
+            case OpenAIChatModel.O1_MINI | OpenAIChatModel.O1_PREVIEW:
+                return ["temperature", "stream"]
+            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
@@ -623,7 +661,11 @@ class OpenAIGPT(LanguageModel):
 
     def get_stream(self) -> bool:
         """Get streaming status"""
-        return self.config.stream and settings.stream
+        return (
+            self.config.stream
+            and settings.stream
+            and self.config.chat_model not in NON_STREAMING_MODELS
+        )
 
     @no_type_check
     def _process_stream_event(
@@ -1421,6 +1463,12 @@ 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
@@ -1457,6 +1505,15 @@ class OpenAIGPT(LanguageModel):
                     tool_choice=tool_choice,
                 )
             )
+        for p in self.unsupported_params():
+            # 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-0.15.2 → langroid-0.16.1}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "langroid"
-version = "0.15.2"
+version = "0.16.1"
 description = "Harness LLMs with Multi-Agent Programming"
 authors = ["Prasad Chalasani <pchalasani@gmail.com>"]
 readme = "README.md"
@@ -53,7 +53,7 @@ trafilatura = "^1.5.0"
 halo = "^0.0.31"
 typer = "^0.9.0"
 colorlog = "^6.7.0"
-openai = "^1.14.0"
+openai = "^1.45.0"
 tiktoken = "^0.7.0"
 pygithub = "^1.58.1"
 pygments = "^2.15.1"