Merge branch 'wip-v0.4' into mdrxy/ollama_v1

libs/core/langchain_core/messages/content_blocks.py

@@ -108,7 +108,6 @@ from uuid import uuid4
 
 from typing_extensions import NotRequired, TypedDict, TypeGuard, get_args, get_origin
 
-
 def _ensure_id(id_val: Optional[str]) -> str:
     """Ensure the ID is a valid string, generating a new UUID if not provided.
 
@@ -314,7 +313,6 @@ class ToolCallChunk(TypedDict):
             + AIMessageChunk(content="", tool_call_chunks=right_chunks)
         ).tool_call_chunks == [ToolCallChunk(name='foo', args='{"a":1}', index=0)]
     """
-
     # TODO: Consider making fields NotRequired[str] in the future.
 
     type: NotRequired[Literal["tool_call_chunk"]]
@@ -586,6 +584,9 @@ class ImageContentBlock(TypedDict):
     """Provider-specific metadata."""
 
 
+class VideoContentBlock(TypedDict):
+    """Video data.
+
 class VideoContentBlock(TypedDict):
     """Video data.
 
@@ -636,7 +637,6 @@ class AudioContentBlock(TypedDict):
     .. note::
         ``create_audio_block`` may also be used as a factory to create an
         ``AudioContentBlock``. Benefits include:
-
         * Automatic ID generation (when not provided)
         * Required arguments strictly validated at creation time
 
@@ -659,6 +659,7 @@ class AudioContentBlock(TypedDict):
     """MIME type of the audio. Required for base64.
 
     `Examples from IANA <https://www.iana.org/assignments/media-types/media-types.xhtml#audio>`__
+
     """
 
     index: NotRequired[int]
@@ -764,6 +765,7 @@ class FileContentBlock(TypedDict):
     """MIME type of the file. Required for base64.
     
     `Examples from IANA <https://www.iana.org/assignments/media-types/media-types.xhtml>`__
+
     """
 
     index: NotRequired[int]
@@ -822,6 +824,48 @@ class NonStandardContentBlock(TypedDict):
 
 # --- Aliases ---
 
+# Future modalities to consider:
+# - 3D models
+# - Tabular data
+
+
+class NonStandardContentBlock(TypedDict):
+    """Provider-specific data.
+
+    This block contains data for which there is not yet a standard type.
+
+    The purpose of this block should be to simply hold a provider-specific payload.
+    If a provider's non-standard output includes reasoning and tool calls, it should be
+    the adapter's job to parse that payload and emit the corresponding standard
+    ReasoningContentBlock and ToolCallContentBlocks.
+
+    .. note::
+        ``create_non_standard_block`` may also be used as a factory to create a
+        ``NonStandardContentBlock``. Benefits include:
+
+        * Automatic ID generation (when not provided)
+        * Required arguments strictly validated at creation time
+
+    """
+
+    type: Literal["non_standard"]
+    """Type of the content block. Used for discrimination."""
+
+    id: NotRequired[str]
+    """Content block identifier. Either:
+
+    - Generated by the provider (e.g., OpenAI's file ID)
+    - Generated by LangChain upon creation (``UUID4`` prefixed with ``'lc_'``))
+    """
+
+    value: dict[str, Any]
+    """Provider-specific data."""
+
+    index: NotRequired[int]
+    """Index of block in aggregate response. Used during streaming."""
+
+
+# --- Aliases ---
 DataContentBlock = Union[
     ImageContentBlock,
     VideoContentBlock,
@@ -955,7 +999,6 @@ def is_invalid_tool_call_block(
     """Type guard to check if a content block is an invalid tool call."""
     return block.get("type") == "invalid_tool_call"
 
-
 def convert_to_openai_image_block(block: dict[str, Any]) -> dict:
     """Convert image content block to format expected by OpenAI Chat Completions API."""
     if "url" in block:

libs/core/langchain_core/tools/base.py

@@ -505,6 +505,7 @@ class ChildTool(BaseTool):
 
     If ``"v0"``, output will be a v0 :class:`~langchain_core.messages.tool.ToolMessage`.
     If ``"v1"``, output will be a v1 :class:`~langchain_core.v1.messages.ToolMessage`.
+
     """
 
     def __init__(self, **kwargs: Any) -> None:

libs/core/langchain_core/v1/messages.py

@@ -368,7 +368,7 @@ class AIMessageChunk(AIMessage):
         raise NotImplementedError(error_msg)
 
     def to_message(self) -> "AIMessage":
-        """Convert this ``AIMessageChunk`` to an AIMessage."""
+        """Convert this ``AIMessageChunk`` to an ``AIMessage``."""
         return AIMessage(
             content=_init_tool_calls(self.content),
             id=self.id,

libs/core/langchain_core/version.py

@@ -1,3 +1,3 @@
 """langchain-core version information and utilities."""
 
-VERSION = "0.3.72"
+VERSION = "0.4.0.dev0"

libs/core/pyproject.toml

@@ -16,7 +16,7 @@ dependencies = [
     "pydantic>=2.7.4",
 ]
 name = "langchain-core"
-version = "0.3.72"
+version = "0.4.0.dev0"
 description = "Building applications with LLMs through composability"
 readme = "README.md"
 

libs/core/tests/unit_tests/messages/test_response_metadata.py.bak

@@ -0,0 +1,361 @@
+"""Unit tests for ResponseMetadata TypedDict."""
+
+from langchain_core.messages.v1 import AIMessage, AIMessageChunk, ResponseMetadata
+
+
+class TestResponseMetadata:
+    """Test the ResponseMetadata TypedDict functionality."""
+
+    def test_response_metadata_basic_fields(self) -> None:
+        """Test ResponseMetadata with basic required fields."""
+        metadata: ResponseMetadata = {
+            "model_provider": "openai",
+            "model_name": "gpt-4",
+        }
+
+        assert metadata.get("model_provider") == "openai"
+        assert metadata.get("model_name") == "gpt-4"
+
+    def test_response_metadata_is_optional(self) -> None:
+        """Test that ResponseMetadata fields are optional due to total=False."""
+        # Should be able to create empty ResponseMetadata
+        metadata: ResponseMetadata = {}
+        assert metadata == {}
+
+        # Should be able to create with just one field
+        metadata_partial: ResponseMetadata = {"model_provider": "anthropic"}
+        assert metadata_partial.get("model_provider") == "anthropic"
+        assert "model_name" not in metadata_partial
+
+    def test_response_metadata_supports_extra_fields(self) -> None:
+        """Test that ResponseMetadata supports provider-specific extra fields."""
+        metadata: ResponseMetadata = {
+            "model_provider": "openai",
+            "model_name": "gpt-4-turbo",
+            # Extra fields should be allowed
+            "usage": {"input_tokens": 100, "output_tokens": 50},
+            "system_fingerprint": "fp_12345",
+            "logprobs": None,
+            "finish_reason": "stop",
+        }
+
+        assert metadata.get("model_provider") == "openai"
+        assert metadata.get("model_name") == "gpt-4-turbo"
+        assert metadata.get("usage") == {"input_tokens": 100, "output_tokens": 50}
+        assert metadata.get("system_fingerprint") == "fp_12345"
+        assert metadata.get("logprobs") is None
+        assert metadata.get("finish_reason") == "stop"
+
+    def test_response_metadata_various_data_types(self) -> None:
+        """Test that ResponseMetadata can store various data types in extra fields."""
+        metadata: ResponseMetadata = {
+            "model_provider": "anthropic",
+            "model_name": "claude-3-sonnet",
+            "string_field": "test_value",  # type: ignore[typeddict-unknown-key]
+            "int_field": 42,  # type: ignore[typeddict-unknown-key]
+            "float_field": 3.14,  # type: ignore[typeddict-unknown-key]
+            "bool_field": True,  # type: ignore[typeddict-unknown-key]
+            "none_field": None,  # type: ignore[typeddict-unknown-key]
+            "list_field": [1, 2, 3, "test"],  # type: ignore[typeddict-unknown-key]
+            "dict_field": {  # type: ignore[typeddict-unknown-key]
+                "nested": {"deeply": "nested_value"}
+            },
+        }
+
+        assert metadata.get("string_field") == "test_value"  # type: ignore[typeddict-item]
+        assert metadata.get("int_field") == 42  # type: ignore[typeddict-item]
+        assert metadata.get("float_field") == 3.14  # type: ignore[typeddict-item]
+        assert metadata.get("bool_field") is True  # type: ignore[typeddict-item]
+        assert metadata.get("none_field") is None  # type: ignore[typeddict-item]
+
+        list_field = metadata.get("list_field")  # type: ignore[typeddict-item]
+        assert isinstance(list_field, list)
+        assert list_field == [1, 2, 3, "test"]
+
+        dict_field = metadata.get("dict_field")  # type: ignore[typeddict-item]
+        assert isinstance(dict_field, dict)
+        nested = dict_field.get("nested")  # type: ignore[union-attr]
+        assert isinstance(nested, dict)
+        assert nested.get("deeply") == "nested_value"  # type: ignore[union-attr]
+
+    def test_response_metadata_can_be_modified(self) -> None:
+        """Test that ResponseMetadata can be modified after creation."""
+        metadata: ResponseMetadata = {
+            "model_provider": "openai",
+            "model_name": "gpt-3.5-turbo",
+        }
+
+        # Modify existing fields
+        metadata["model_name"] = "gpt-4"
+        assert metadata.get("model_name") == "gpt-4"
+
+        # Add new fields
+        metadata["request_id"] = "req_12345"  # type: ignore[typeddict-unknown-key]
+        assert metadata.get("request_id") == "req_12345"  # type: ignore[typeddict-item]
+
+        # Modify nested structures
+        metadata["usage"] = {"input_tokens": 10}  # type: ignore[typeddict-unknown-key]
+        metadata["usage"]["output_tokens"] = 20  # type: ignore[typeddict-item]
+
+        usage = metadata.get("usage")  # type: ignore[typeddict-item]
+        assert isinstance(usage, dict)
+        assert usage.get("input_tokens") == 10  # type: ignore[union-attr]
+        assert usage.get("output_tokens") == 20  # type: ignore[union-attr]
+
+    def test_response_metadata_provider_specific_examples(self) -> None:
+        """Test ResponseMetadata with realistic provider-specific examples."""
+        # OpenAI-style metadata
+        openai_metadata: ResponseMetadata = {
+            "model_provider": "openai",
+            "model_name": "gpt-4-turbo-2024-04-09",
+            "usage": {  # type: ignore[typeddict-unknown-key]
+                "prompt_tokens": 50,
+                "completion_tokens": 25,
+                "total_tokens": 75,
+            },
+            "system_fingerprint": "fp_abc123",  # type: ignore[typeddict-unknown-key]
+            "created": 1234567890,  # type: ignore[typeddict-unknown-key]
+            "logprobs": None,  # type: ignore[typeddict-unknown-key]
+            "finish_reason": "stop",  # type: ignore[typeddict-unknown-key]
+        }
+
+        assert openai_metadata.get("model_provider") == "openai"
+        assert openai_metadata.get("system_fingerprint") == "fp_abc123"  # type: ignore[typeddict-item]
+
+        # Anthropic-style metadata
+        anthropic_metadata: ResponseMetadata = {
+            "model_provider": "anthropic",
+            "model_name": "claude-3-sonnet-20240229",
+            "usage": {  # type: ignore[typeddict-unknown-key]
+                "input_tokens": 75,
+                "output_tokens": 30,
+            },
+            "stop_reason": "end_turn",  # type: ignore[typeddict-unknown-key]
+            "stop_sequence": None,  # type: ignore[typeddict-unknown-key]
+        }
+
+        assert anthropic_metadata.get("model_provider") == "anthropic"
+        assert anthropic_metadata.get("stop_reason") == "end_turn"  # type: ignore[typeddict-item]
+
+        # Custom provider metadata
+        custom_metadata: ResponseMetadata = {
+            "model_provider": "custom_llm_service",
+            "model_name": "custom-model-v1",
+            "service_tier": "premium",  # type: ignore[typeddict-unknown-key]
+            "rate_limit_info": {  # type: ignore[typeddict-unknown-key]
+                "requests_remaining": 100,
+                "reset_time": "2024-01-01T00:00:00Z",
+            },
+            "response_time_ms": 1250,  # type: ignore[typeddict-unknown-key]
+        }
+
+        assert custom_metadata.get("service_tier") == "premium"  # type: ignore[typeddict-item]
+        rate_limit = custom_metadata.get("rate_limit_info")  # type: ignore[typeddict-item]
+        assert isinstance(rate_limit, dict)
+        assert rate_limit.get("requests_remaining") == 100  # type: ignore[union-attr]
+
+
+class TestResponseMetadataWithAIMessages:
+    """Test ResponseMetadata integration with AI message classes."""
+
+    def test_ai_message_with_response_metadata(self) -> None:
+        """Test AIMessage with ResponseMetadata."""
+        metadata: ResponseMetadata = {
+            "model_provider": "openai",
+            "model_name": "gpt-4",
+            "usage": {"input_tokens": 10, "output_tokens": 5},  # type: ignore[typeddict-unknown-key]
+        }
+
+        message = AIMessage(content="Hello, world!", response_metadata=metadata)
+
+        assert message.response_metadata == metadata
+        assert message.response_metadata.get("model_provider") == "openai"
+        assert message.response_metadata.get("model_name") == "gpt-4"
+
+        usage = message.response_metadata.get("usage")  # type: ignore[typeddict-item]
+        assert isinstance(usage, dict)
+        assert usage.get("input_tokens") == 10  # type: ignore[union-attr]
+
+    def test_ai_message_chunk_with_response_metadata(self) -> None:
+        """Test AIMessageChunk with ResponseMetadata."""
+        metadata: ResponseMetadata = {
+            "model_provider": "anthropic",
+            "model_name": "claude-3-sonnet",
+            "stream_id": "stream_12345",  # type: ignore[typeddict-unknown-key]
+        }
+
+        chunk = AIMessageChunk(content="Hello", response_metadata=metadata)
+
+        assert chunk.response_metadata == metadata
+        assert chunk.response_metadata.get("stream_id") == "stream_12345"  # type: ignore[typeddict-item]
+
+    def test_ai_message_default_empty_response_metadata(self) -> None:
+        """Test that AIMessage creates empty ResponseMetadata by default."""
+        message = AIMessage(content="Test message")
+
+        # Should have empty dict as default
+        assert message.response_metadata == {}
+        assert isinstance(message.response_metadata, dict)
+
+    def test_ai_message_chunk_default_empty_response_metadata(self) -> None:
+        """Test that AIMessageChunk creates empty ResponseMetadata by default."""
+        chunk = AIMessageChunk(content="Test chunk")
+
+        # Should have empty dict as default
+        assert chunk.response_metadata == {}
+        assert isinstance(chunk.response_metadata, dict)
+
+    def test_response_metadata_merging_in_chunks(self) -> None:
+        """Test that ResponseMetadata is properly merged when adding AIMessageChunks."""
+        metadata1: ResponseMetadata = {
+            "model_provider": "openai",
+            "model_name": "gpt-4",
+            "request_id": "req_123",  # type: ignore[typeddict-unknown-key]
+            "usage": {"input_tokens": 10},  # type: ignore[typeddict-unknown-key]
+        }
+
+        metadata2: ResponseMetadata = {
+            "stream_chunk": 1,  # type: ignore[typeddict-unknown-key]
+            "usage": {"output_tokens": 5},  # type: ignore[typeddict-unknown-key]
+        }
+
+        chunk1 = AIMessageChunk(content="Hello ", response_metadata=metadata1)
+        chunk2 = AIMessageChunk(content="world!", response_metadata=metadata2)
+
+        merged = chunk1 + chunk2
+
+        # Should have merged response_metadata
+        assert merged.response_metadata.get("model_provider") == "openai"
+        assert merged.response_metadata.get("model_name") == "gpt-4"
+        assert merged.response_metadata.get("request_id") == "req_123"  # type: ignore[typeddict-item]
+        assert merged.response_metadata.get("stream_chunk") == 1  # type: ignore[typeddict-item]
+
+        # Usage should be merged (from merge_dicts behavior)
+        merged_usage = merged.response_metadata.get("usage")  # type: ignore[typeddict-item]
+        assert isinstance(merged_usage, dict)
+        assert merged_usage.get("input_tokens") == 10  # type: ignore[union-attr]
+        assert merged_usage.get("output_tokens") == 5  # type: ignore[union-attr]
+
+    def test_response_metadata_modification_after_message_creation(self) -> None:
+        """Test that ResponseMetadata can be modified after message creation."""
+        message = AIMessage(
+            content="Initial message",
+            response_metadata={"model_provider": "openai", "model_name": "gpt-3.5"},
+        )
+
+        # Modify existing field
+        message.response_metadata["model_name"] = "gpt-4"
+        assert message.response_metadata.get("model_name") == "gpt-4"
+
+        # Add new field
+        message.response_metadata["finish_reason"] = "stop"  # type: ignore[typeddict-unknown-key]
+        assert message.response_metadata.get("finish_reason") == "stop"  # type: ignore[typeddict-item]
+
+    def test_response_metadata_with_none_values(self) -> None:
+        """Test ResponseMetadata handling of None values."""
+        metadata: ResponseMetadata = {
+            "model_provider": "openai",
+            "model_name": "gpt-4",
+            "system_fingerprint": None,  # type: ignore[typeddict-unknown-key]
+            "logprobs": None,  # type: ignore[typeddict-unknown-key]
+        }
+
+        message = AIMessage(content="Test", response_metadata=metadata)
+
+        assert message.response_metadata.get("system_fingerprint") is None  # type: ignore[typeddict-item]
+        assert message.response_metadata.get("logprobs") is None  # type: ignore[typeddict-item]
+        assert "system_fingerprint" in message.response_metadata
+        assert "logprobs" in message.response_metadata
+
+
+class TestResponseMetadataEdgeCases:
+    """Test edge cases and error conditions for ResponseMetadata."""
+
+    def test_response_metadata_with_complex_nested_structures(self) -> None:
+        """Test ResponseMetadata with deeply nested and complex structures."""
+        metadata: ResponseMetadata = {
+            "model_provider": "custom",
+            "model_name": "complex-model",
+            "complex_data": {  # type: ignore[typeddict-unknown-key]
+                "level1": {
+                    "level2": {
+                        "level3": {
+                            "deeply_nested": "value",
+                            "array": [
+                                {"item": 1, "metadata": {"nested": True}},
+                                {"item": 2, "metadata": {"nested": False}},
+                            ],
+                        }
+                    }
+                }
+            },
+        }
+
+        complex_data = metadata.get("complex_data")  # type: ignore[typeddict-item]
+        assert isinstance(complex_data, dict)
+        level1 = complex_data.get("level1")  # type: ignore[union-attr]
+        assert isinstance(level1, dict)
+        level2 = level1.get("level2")  # type: ignore[union-attr]
+        assert isinstance(level2, dict)
+        level3 = level2.get("level3")  # type: ignore[union-attr]
+        assert isinstance(level3, dict)
+
+        assert level3.get("deeply_nested") == "value"  # type: ignore[union-attr]
+        array = level3.get("array")  # type: ignore[union-attr]
+        assert isinstance(array, list)
+        assert len(array) == 2  # type: ignore[arg-type]
+        assert array[0]["item"] == 1  # type: ignore[index, typeddict-item]
+        assert array[0]["metadata"]["nested"] is True  # type: ignore[index, typeddict-item]
+
+    def test_response_metadata_large_data(self) -> None:
+        """Test ResponseMetadata with large amounts of data."""
+        # Create metadata with many fields
+        large_metadata: ResponseMetadata = {
+            "model_provider": "test_provider",
+            "model_name": "test_model",
+        }
+
+        # Add 100 extra fields
+        for i in range(100):
+            large_metadata[f"field_{i}"] = f"value_{i}"  # type: ignore[literal-required]
+
+        message = AIMessage(content="Test", response_metadata=large_metadata)
+
+        # Verify all fields are accessible
+        assert message.response_metadata.get("model_provider") == "test_provider"
+        for i in range(100):
+            assert message.response_metadata.get(f"field_{i}") == f"value_{i}"  # type: ignore[typeddict-item]
+
+    def test_response_metadata_empty_vs_none(self) -> None:
+        """Test the difference between empty ResponseMetadata and None."""
+        # Message with empty metadata
+        message_empty = AIMessage(content="Test", response_metadata={})
+        assert message_empty.response_metadata == {}
+        assert isinstance(message_empty.response_metadata, dict)
+
+        # Message with None metadata (should become empty dict)
+        message_none = AIMessage(content="Test", response_metadata=None)
+        assert message_none.response_metadata == {}
+        assert isinstance(message_none.response_metadata, dict)
+
+        # Default message (no metadata specified)
+        message_default = AIMessage(content="Test")
+        assert message_default.response_metadata == {}
+        assert isinstance(message_default.response_metadata, dict)
+
+    def test_response_metadata_preserves_original_dict_type(self) -> None:
+        """Test that ResponseMetadata preserves the original dict when passed."""
+        original_dict = {
+            "model_provider": "openai",
+            "model_name": "gpt-4",
+            "custom_field": "custom_value",
+        }
+
+        message = AIMessage(content="Test", response_metadata=original_dict)
+
+        # Should be the same dict object
+        assert message.response_metadata is original_dict
+
+        # Modifications to the message's response_metadata should affect original
+        message.response_metadata["new_field"] = "new_value"  # type: ignore[typeddict-unknown-key]
+        assert original_dict.get("new_field") == "new_value"  # type: ignore[typeddict-item]

libs/core/uv.lock

@@ -987,7 +987,7 @@ wheels = [
 
 [[package]]
 name = "langchain-core"
-version = "0.3.72"
+version = "0.4.0.dev0"
 source = { editable = "." }
 dependencies = [
     { name = "jsonpatch" },

libs/langchain/pyproject.toml

@@ -7,7 +7,7 @@ authors = []
 license = { text = "MIT" }
 requires-python = ">=3.9, <4.0"
 dependencies = [
-    "langchain-core<1.0.0,>=0.3.72",
+    "langchain-core<1.0.0,>=0.4.0.dev0",
     "langchain-text-splitters<1.0.0,>=0.3.9",
     "langsmith>=0.1.17",
     "pydantic<3.0.0,>=2.7.4",
@@ -17,7 +17,7 @@ dependencies = [
     "async-timeout<5.0.0,>=4.0.0; python_version < \"3.11\"",
 ]
 name = "langchain"
-version = "0.3.27"
+version = "0.4.0.dev0"
 description = "Building applications with LLMs through composability"
 readme = "README.md"
 
@@ -25,9 +25,9 @@ readme = "README.md"
 community = ["langchain-community"]
 anthropic = ["langchain-anthropic"]
 openai = ["langchain-openai"]
-azure-ai = ["langchain-azure-ai"]
+# azure-ai = ["langchain-azure-ai"]
-cohere = ["langchain-cohere"]
+# cohere = ["langchain-cohere"]
-google-vertexai = ["langchain-google-vertexai"]
+# google-vertexai = ["langchain-google-vertexai"]
 google-genai = ["langchain-google-genai"]
 fireworks = ["langchain-fireworks"]
 ollama = ["langchain-ollama"]
@@ -35,9 +35,9 @@ together = ["langchain-together"]
 mistralai = ["langchain-mistralai"]
 huggingface = ["langchain-huggingface"]
 groq = ["langchain-groq"]
-aws = ["langchain-aws"]
+# aws = ["langchain-aws"]
 deepseek = ["langchain-deepseek"]
-xai = ["langchain-xai"]
+# xai = ["langchain-xai"]
 perplexity = ["langchain-perplexity"]
 
 [project.urls]

libs/partners/openai/pyproject.toml

@@ -7,12 +7,12 @@ authors = []
 license = { text = "MIT" }
 requires-python = ">=3.9"
 dependencies = [
-    "langchain-core<1.0.0,>=0.3.68",
+    "langchain-core<1.0.0,>=0.4.0.dev0",
     "openai<2.0.0,>=1.86.0",
     "tiktoken<1,>=0.7",
 ]
 name = "langchain-openai"
-version = "0.3.28"
+version = "0.4.0.dev0"
 description = "An integration package connecting OpenAI and LangChain"
 readme = "README.md"
 

libs/partners/openai/uv.lock

@@ -480,7 +480,7 @@ wheels = [
 
 [[package]]
 name = "langchain-core"
-version = "0.3.72"
+version = "0.4.0.dev0"
 source = { editable = "../../core" }
 dependencies = [
     { name = "jsonpatch" },
@@ -538,7 +538,7 @@ typing = [
 
 [[package]]
 name = "langchain-openai"
-version = "0.3.28"
+version = "0.4.0.dev0"
 source = { editable = "." }
 dependencies = [
     { name = "langchain-core" },