x

Llama-3.1 started failing consistently with
> groq.BadRequestError: Error code: 400 - ***'error': ***'message':
"Failed to call a function. Please adjust your prompt. See
'failed_generation' for more details.", 'type': 'invalid_request_error',
'code': 'tool_use_failed', 'failed_generation':
'<function=brave_search>***"query": "Hello!"***</function>'***

.github/workflows/codspeed.yml

@@ -8,10 +8,8 @@ on:
   workflow_dispatch:
 
 env:
-  AZURE_OPENAI_API_VERSION: ${{ secrets.AZURE_OPENAI_API_VERSION }}
-  AZURE_OPENAI_API_BASE: ${{ secrets.AZURE_OPENAI_API_BASE }}
-  AZURE_OPENAI_CHAT_DEPLOYMENT_NAME: ${{ secrets.AZURE_OPENAI_CHAT_DEPLOYMENT_NAME }}
-  AZURE_OPENAI_LEGACY_CHAT_DEPLOYMENT_NAME: ${{ secrets.AZURE_OPENAI_LEGACY_CHAT_DEPLOYMENT_NAME }}
+  AZURE_OPENAI_CHAT_DEPLOYMENT_NAME: foo
+  AZURE_OPENAI_LEGACY_CHAT_DEPLOYMENT_NAME: foo
   DEEPSEEK_API_KEY: foo
   FIREWORKS_API_KEY: foo
 
@@ -62,4 +60,3 @@ jobs:
               uv run --no-sync pytest ./tests/ --codspeed
             fi
           mode: ${{ matrix.mode || 'instrumentation' }}
-

docs/docs/how_to/assign.ipynb

@@ -157,7 +157,7 @@
     "\n",
     "## Next steps\n",
     "\n",
-    "Now you've learned how to pass data through your chains to help to help format the data flowing through your chains.\n",
+    "Now you've learned how to pass data through your chains to help format the data flowing through your chains.\n",
     "\n",
     "To learn more, see the other how-to guides on runnables in this section."
    ]

docs/docs/integrations/chat/fireworks.ipynb

@@ -17,7 +17,7 @@
       "source": [
         "# ChatFireworks\n",
         "\n",
-        "This doc help you get started with Fireworks AI [chat models](/docs/concepts/chat_models). For detailed documentation of all ChatFireworks features and configurations head to the [API reference](https://python.langchain.com/api_reference/fireworks/chat_models/langchain_fireworks.chat_models.ChatFireworks.html).\n",
+        "This doc helps you get started with Fireworks AI [chat models](/docs/concepts/chat_models). For detailed documentation of all ChatFireworks features and configurations head to the [API reference](https://python.langchain.com/api_reference/fireworks/chat_models/langchain_fireworks.chat_models.ChatFireworks.html).\n",
         "\n",
         "Fireworks AI is an AI inference platform to run and customize models. For a list of all models served by Fireworks see the [Fireworks docs](https://fireworks.ai/models).\n",
         "\n",
@@ -39,7 +39,7 @@
         "\n",
         "### Credentials\n",
         "\n",
-        "Head to (ttps://fireworks.ai/login to sign up to Fireworks and generate an API key. Once you've done this set the FIREWORKS_API_KEY environment variable:"
+        "Head to (https://fireworks.ai/login to sign up to Fireworks and generate an API key. Once you've done this set the FIREWORKS_API_KEY environment variable:"
       ]
     },
     {

docs/docs/integrations/chat/kinetica.ipynb

@@ -61,7 +61,7 @@
     "# Install Langchain community and core packages\n",
     "%pip install --upgrade --quiet langchain-core langchain-community\n",
     "\n",
-    "# Install Kineitca DB connection package\n",
+    "# Install Kinetica DB connection package\n",
     "%pip install --upgrade --quiet 'gpudb>=7.2.0.8' typeguard pandas tqdm\n",
     "\n",
     "# Install packages needed for this tutorial\n",

docs/docs/integrations/chat/nvidia_ai_endpoints.ipynb

@@ -318,7 +318,7 @@
       "source": [
         "### Code Generation\n",
         "\n",
-        "These models accept the same arguments and input structure as regular chat models, but they tend to perform better on code-genreation and structured code tasks. An example of this is `meta/codellama-70b`."
+        "These models accept the same arguments and input structure as regular chat models, but they tend to perform better on code-generation and structured code tasks. An example of this is `meta/codellama-70b`."
       ]
     },
     {

docs/docs/integrations/chat/reka.ipynb

@@ -36,7 +36,7 @@
     "\n",
     "## Setup\n",
     "\n",
-    "To access Reka models you'll need to create an Reka developer account, get an API key, and install the `langchain_community` integration package and the reka python package via 'pip install reka-api'.\n",
+    "To access Reka models you'll need to create a Reka developer account, get an API key, and install the `langchain_community` integration package and the reka python package via 'pip install reka-api'.\n",
     "\n",
     "### Credentials\n",
     "\n",
@@ -280,7 +280,7 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "Use use with tavtly api search"
+    "Use with Tavily api search"
    ]
   },
   {

docs/docs/integrations/chat/writer.ipynb

@@ -303,7 +303,7 @@
    "source": [
     "### A note on tool binding\n",
     "\n",
-    "The `ChatWriter.bind_tools()` method does not create new instance with bound tools, but stores the received `tools` and `tool_choice` in the initial class instance attributes to pass them as parameters during the Palmyra LLM call while using `ChatWriter` invocation. This approach allows the support of different tool types, e.g. `function` and `graph`. `Graph` is one of the remotely called Writer Palmyra tools. For further information visit our [docs](https://dev.writer.com/api-guides/knowledge-graph#knowledge-graph). \n",
+    "The `ChatWriter.bind_tools()` method does not create a new instance with bound tools, but stores the received `tools` and `tool_choice` in the initial class instance attributes to pass them as parameters during the Palmyra LLM call while using `ChatWriter` invocation. This approach allows the support of different tool types, e.g. `function` and `graph`. `Graph` is one of the remotely called Writer Palmyra tools. For further information, visit our [docs](https://dev.writer.com/api-guides/knowledge-graph#knowledge-graph). \n",
     "\n",
     "For more information about tool usage in LangChain, visit the [LangChain tool calling documentation](https://python.langchain.com/docs/concepts/tool_calling/)."
    ]
@@ -373,7 +373,7 @@
    "source": [
     "## Prompt templates\n",
     "\n",
-    "[Prompt templates](https://python.langchain.com/docs/concepts/prompt_templates/) help to translate user input and parameters into instructions for a language model. You can use `ChatWriter` with a prompt templates like so:\n"
+    "[Prompt templates](https://python.langchain.com/docs/concepts/prompt_templates/) help to translate user input and parameters into instructions for a language model. You can use `ChatWriter` with a prompt template like so:\n"
    ]
   },
   {
@@ -411,7 +411,7 @@
    "metadata": {},
    "source": [
     "## API reference\n",
-    "For detailed documentation of all ChatWriter features and configurations head to the [API reference](https://python.langchain.com/api_reference/writer/chat_models/langchain_writer.chat_models.ChatWriter.html#langchain_writer.chat_models.ChatWriter).\n",
+    "For detailed documentation of all ChatWriter features and configurations, head to the [API reference](https://python.langchain.com/api_reference/writer/chat_models/langchain_writer.chat_models.ChatWriter.html#langchain_writer.chat_models.ChatWriter).\n",
     "\n",
     "## Additional resources\n",
     "You can find information about Writer's models (including costs, context windows, and supported input types) and tools in the [Writer docs](https://dev.writer.com/home)."

docs/docs/integrations/llms/sambanovacloud.ipynb

@@ -74,14 +74,14 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 2,
+   "execution_count": null,
    "metadata": {},
    "outputs": [],
    "source": [
     "from langchain_community.llms.sambanova import SambaNovaCloud\n",
     "\n",
     "llm = SambaNovaCloud(\n",
-    "    model=\"Meta-Llama-3.1-70B-Instruct\",\n",
+    "    model=\"Meta-Llama-3.3-70B-Instruct\",\n",
     "    max_tokens_to_generate=1000,\n",
     "    temperature=0.01,\n",
     "    # top_k = 50,\n",

docs/docs/integrations/providers/sambanova.ipynb

@@ -130,7 +130,7 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "For a more detailed walkthrough of the SambaStudioEmbeddings component, see [this notebook](https://python.langchain.com/docs/integrations/text_embedding/sambanova/)"
+    "For a more detailed walkthrough of the SambaNovaCloudEmbeddings component, see [this notebook](https://python.langchain.com/docs/integrations/text_embedding/sambanova/)"
    ]
   },
   {

libs/partners/groq/tests/integration_tests/test_standard.py

@@ -29,14 +29,10 @@ class BaseTestGroq(ChatModelIntegrationTests):
         return True
 
 
-class TestGroqLlama(BaseTestGroq):
+class TestGroqGemma(BaseTestGroq):
     @property
     def chat_model_params(self) -> dict:
-        return {
-            "model": "llama-3.1-8b-instant",
-            "temperature": 0,
-            "rate_limiter": rate_limiter,
-        }
+        return {"model": "gemma2-9b-it", "rate_limiter": rate_limiter}
 
     @property
     def supports_json_mode(self) -> bool:

libs/partners/huggingface/langchain_huggingface/chat_models/huggingface.py

@@ -775,12 +775,12 @@ class ChatHuggingFace(BaseChatModel):
         elif _is_huggingface_pipeline(self.llm):
             from transformers import AutoTokenizer  # type: ignore[import]
 
+            self.model_id = self.model_id or self.llm.model_id
             self.tokenizer = (
                 AutoTokenizer.from_pretrained(self.model_id)
                 if self.tokenizer is None
                 else self.tokenizer
             )
-            self.model_id = self.llm.model_id
             return
         elif _is_huggingface_endpoint(self.llm):
             self.model_id = self.llm.repo_id or self.llm.model

libs/partners/openai/langchain_openai/chat_models/base.py

@@ -118,6 +118,15 @@ global_ssl_context = ssl.create_default_context(cafile=certifi.where())
 
 _FUNCTION_CALL_IDS_MAP_KEY = "__openai_function_call_ids__"
 
+WellKnownTools = (
+    "file_search",
+    "web_search_preview",
+    "computer_use_preview",
+    "code_interpreter",
+    "mcp",
+    "image_generation",
+)
+
 
 def _convert_dict_to_message(_dict: Mapping[str, Any]) -> BaseMessage:
     """Convert a dictionary to a LangChain message.
@@ -1487,13 +1496,7 @@ class BaseChatOpenAI(BaseChatModel):
                         "type": "function",
                         "function": {"name": tool_choice},
                     }
-                elif tool_choice in (
-                    "file_search",
-                    "web_search_preview",
-                    "computer_use_preview",
-                    "code_interpreter",
-                    "mcp",
-                ):
+                elif tool_choice in WellKnownTools:
                     tool_choice = {"type": tool_choice}
                 # 'any' is not natively supported by OpenAI API.
                 # We support 'any' since other models use this instead of 'required'.
@@ -3050,6 +3053,13 @@ def _construct_responses_api_payload(
                 new_tools.append({"type": "function", **tool["function"]})
             else:
                 new_tools.append(tool)
+
+            if tool["type"] == "image_generation" and "partial_images" in tool:
+                raise NotImplementedError(
+                    "Partial image generation is not yet supported "
+                    "via the LangChain ChatOpenAI client. Please "
+                    "drop the 'partial_images' key from the image_generation tool."
+                )
         payload["tools"] = new_tools
     if tool_choice := payload.pop("tool_choice", None):
         # chat api: {"type": "function", "function": {"name": "..."}}
@@ -3139,6 +3149,7 @@ def _pop_summary_index_from_reasoning(reasoning: dict) -> dict:
 
 
 def _construct_responses_api_input(messages: Sequence[BaseMessage]) -> list:
+    """Construct the input for the OpenAI Responses API."""
     input_ = []
     for lc_msg in messages:
         msg = _convert_message_to_dict(lc_msg)
@@ -3191,6 +3202,7 @@ def _construct_responses_api_input(messages: Sequence[BaseMessage]) -> list:
             computer_calls = []
             code_interpreter_calls = []
             mcp_calls = []
+            image_generation_calls = []
             tool_outputs = lc_msg.additional_kwargs.get("tool_outputs", [])
             for tool_output in tool_outputs:
                 if tool_output.get("type") == "computer_call":
@@ -3199,10 +3211,22 @@ def _construct_responses_api_input(messages: Sequence[BaseMessage]) -> list:
                     code_interpreter_calls.append(tool_output)
                 elif tool_output.get("type") == "mcp_call":
                     mcp_calls.append(tool_output)
+                elif tool_output.get("type") == "image_generation":
+                    image_generation_calls.append(tool_output)
                 else:
                     pass
             input_.extend(code_interpreter_calls)
             input_.extend(mcp_calls)
+
+            # A previous image generation call can be referenced by ID
+
+            input_.extend(
+                [
+                    {"type": "image_generation", "id": image_generation_call["id"]}
+                    for image_generation_call in image_generation_calls
+                ]
+            )
+
             msg["content"] = msg.get("content") or []
             if lc_msg.additional_kwargs.get("refusal"):
                 if isinstance(msg["content"], str):
@@ -3489,6 +3513,7 @@ def _convert_responses_chunk_to_generation_chunk(
         "mcp_call",
         "mcp_list_tools",
         "mcp_approval_request",
+        "image_generation_call",
     ):
         additional_kwargs["tool_outputs"] = [
             chunk.item.model_dump(exclude_none=True, mode="json")
@@ -3516,6 +3541,9 @@ def _convert_responses_chunk_to_generation_chunk(
                 {"index": chunk.summary_index, "type": "summary_text", "text": ""}
             ]
         }
+    elif chunk.type == "response.image_generation_call.partial_image":
+        # Partial images are not supported yet.
+        pass
     elif chunk.type == "response.reasoning_summary_text.delta":
         additional_kwargs["reasoning"] = {
             "summary": [

libs/partners/openai/pyproject.toml

@@ -7,7 +7,7 @@ authors = []
 license = { text = "MIT" }
 requires-python = ">=3.9"
 dependencies = [
-    "langchain-core<1.0.0,>=0.3.61",
+    "langchain-core<1.0.0,>=0.3.63",
     "openai<2.0.0,>=1.68.2",
     "tiktoken<1,>=0.7",
 ]

libs/partners/openai/tests/cassettes/TestAzureOpenAIStandardLegacy_test_stream_time.yaml.gz

@@ -1 +0,0 @@
-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

libs/partners/openai/tests/integration_tests/chat_models/test_azure_standard.py

@@ -38,10 +38,6 @@ class TestAzureOpenAIStandard(ChatModelIntegrationTests):
     def supports_json_mode(self) -> bool:
         return True
 
-    @property
-    def enable_vcr_tests(self) -> bool:
-        return True
-
 
 class TestAzureOpenAIStandardLegacy(ChatModelIntegrationTests):
     """Test a legacy model."""
@@ -62,7 +58,3 @@ class TestAzureOpenAIStandardLegacy(ChatModelIntegrationTests):
     @property
     def structured_output_kwargs(self) -> dict:
         return {"method": "function_calling"}
-
-    @property
-    def enable_vcr_tests(self) -> bool:
-        return True

libs/partners/openai/tests/integration_tests/chat_models/test_responses_api.py

@@ -12,6 +12,7 @@ from langchain_core.messages import (
     BaseMessage,
     BaseMessageChunk,
     HumanMessage,
+    MessageLikeRepresentation,
 )
 from pydantic import BaseModel
 from typing_extensions import TypedDict
@@ -452,3 +453,126 @@ def test_mcp_builtin() -> None:
     _ = llm_with_tools.invoke(
         [approval_message], previous_response_id=response.response_metadata["id"]
     )
+
+
+@pytest.mark.skip(reason="Re-enable once VCR is properly set up.")
+@pytest.mark.vcr()
+def test_image_generation_streaming() -> None:
+    """Test image generation streaming."""
+    llm = ChatOpenAI(model="gpt-4.1", use_responses_api=True)
+    # Test invocation
+    tool = {
+        "type": "image_generation",
+        # For testing purposes let's keep the quality low, so the test runs faster.
+        "quality": "low",
+    }
+    llm_with_tools = llm.bind_tools([tool])
+    response = llm_with_tools.invoke("Make a picture of a fuzzy cat")
+    _check_response(response)
+    tool_output = response.additional_kwargs["tool_outputs"][0]
+
+    # Example tool output for an image
+    # {
+    #     "background": "opaque",
+    #     "id": "ig_683716a8ddf0819888572b20621c7ae4029ec8c11f8dacf8",
+    #     "output_format": "png",
+    #     "quality": "high",
+    #     "revised_prompt": "A fluffy, fuzzy cat sitting calmly, with soft fur, bright "
+    #     "eyes, and a cute, friendly expression. The background is "
+    #     "simple and light to emphasize the cat's texture and "
+    #     "fluffiness.",
+    #     "size": "1024x1024",
+    #     "status": "completed",
+    #     "type": "image_generation_call",
+    #     "result": # base64 encode image data
+    # }
+
+    expected_keys = {
+        "id",
+        "background",
+        "output_format",
+        "quality",
+        "result",
+        "revised_prompt",
+        "size",
+        "status",
+        "type",
+    }
+
+    assert set(tool_output.keys()).issubset(expected_keys)
+    llm = ChatOpenAI(model="gpt-4.1", use_responses_api=True)
+    full: Optional[BaseMessageChunk] = None
+    tool = {"type": "image_generation", "quality": "low"}
+    for chunk in llm.stream("Make a picture of a fuzzy cat", tools=[tool]):
+        assert isinstance(chunk, AIMessageChunk)
+        full = chunk if full is None else full + chunk
+    complete_ai_message = cast(AIMessageChunk, full)
+    # At the moment, the streaming API does not pick up annotations fully.
+    # So the following check is commented out.
+    # _check_response(complete_ai_message)
+    tool_output = complete_ai_message.additional_kwargs["tool_outputs"][0]
+    assert set(tool_output.keys()).issubset(expected_keys)
+
+
+def test_image_generation_multi_turn() -> None:
+    """Test multi-turn editing of image generation by passing in history."""
+    # Test multi-turn
+    llm = ChatOpenAI(model="gpt-4.1", use_responses_api=True)
+    # Test invocation
+    tool = {
+        "type": "image_generation",
+        # For testing purposes let's keep the quality low, so the test runs faster.
+        "quality": "low",
+    }
+    llm_with_tools = llm.bind_tools([tool])
+
+    chat_history: list[MessageLikeRepresentation] = [
+        {"role": "user", "content": "Make a picture of a fuzzy cat"}
+    ]
+    ai_message = llm_with_tools.invoke(chat_history)
+    _check_response(ai_message)
+    tool_output = ai_message.additional_kwargs["tool_outputs"][0]
+
+    # Example tool output for an image
+    # {
+    #     "background": "opaque",
+    #     "id": "ig_683716a8ddf0819888572b20621c7ae4029ec8c11f8dacf8",
+    #     "output_format": "png",
+    #     "quality": "high",
+    #     "revised_prompt": "A fluffy, fuzzy cat sitting calmly, with soft fur, bright "
+    #     "eyes, and a cute, friendly expression. The background is "
+    #     "simple and light to emphasize the cat's texture and "
+    #     "fluffiness.",
+    #     "size": "1024x1024",
+    #     "status": "completed",
+    #     "type": "image_generation_call",
+    #     "result": # base64 encode image data
+    # }
+
+    expected_keys = {
+        "id",
+        "background",
+        "output_format",
+        "quality",
+        "result",
+        "revised_prompt",
+        "size",
+        "status",
+        "type",
+    }
+
+    assert set(tool_output.keys()).issubset(expected_keys)
+
+    chat_history.extend(
+        [
+            # AI message with tool output
+            ai_message,
+            # New request
+            {"role": "user", "content": "Now make it realistic."},
+        ]
+    )
+
+    ai_message2 = llm_with_tools.invoke(chat_history)
+    _check_response(ai_message2)
+    tool_output2 = ai_message2.additional_kwargs["tool_outputs"][0]
+    assert set(tool_output2.keys()).issubset(expected_keys)