feat(anthropic): v1 support (#32623)

Author: ccurme

libs/core/langchain_core/language_models/_utils.py

@@ -212,7 +212,7 @@ def _normalize_messages(
     }
 
     """
-    from langchain_core.messages.block_translators.langchain import (
+    from langchain_core.messages.block_translators.langchain_v0 import (
         _convert_legacy_v0_content_block_to_v1,
         _convert_openai_format_to_data_block,
     )

libs/core/langchain_core/language_models/chat_models.py

@@ -124,7 +124,7 @@ def _format_for_tracing(messages: list[BaseMessage]) -> list[BaseMessage]:
                     if (
                         block.get("type") == "image"
                         and is_data_content_block(block)
-                        and block.get("source_type") != "id"
+                        and not ("file_id" in block or block.get("source_type") == "id")
                     ):
                         if message_to_trace is message:
                             # Shallow copy

libs/core/langchain_core/messages/ai.py

@@ -231,7 +231,10 @@ def content_blocks(self) -> list[types.ContentBlock]:
 
             translator = get_translator(model_provider)
             if translator:
-                return translator["translate_content"](self)
+                try:
+                    return translator["translate_content_chunk"](self)
+                except NotImplementedError:
+                    pass
 
         # Otherwise, use best-effort parsing
         blocks = super().content_blocks
@@ -380,7 +383,10 @@ def content_blocks(self) -> list[types.ContentBlock]:
 
             translator = get_translator(model_provider)
             if translator:
-                return translator["translate_content_chunk"](self)
+                try:
+                    return translator["translate_content_chunk"](self)
+                except NotImplementedError:
+                    pass
 
         # Otherwise, use best-effort parsing
         blocks = super().content_blocks

libs/core/langchain_core/messages/base.py

@@ -7,21 +7,14 @@
 from pydantic import ConfigDict, Field
 
 from langchain_core.load.serializable import Serializable
-from langchain_core.messages.block_translators.langchain import (
-    _convert_legacy_v0_content_block_to_v1,
-    _convert_v0_multimodal_input_to_v1,
-)
-from langchain_core.messages.block_translators.openai import (
-    _convert_to_v1_from_chat_completions_input,
-)
+from langchain_core.messages import content as types
 from langchain_core.utils import get_bolded_text
 from langchain_core.utils._merge import merge_dicts, merge_lists
 from langchain_core.utils.interactive_env import is_interactive_env
 
 if TYPE_CHECKING:
     from collections.abc import Sequence
 
-    from langchain_core.messages import content as types
     from langchain_core.prompts.chat import ChatPromptTemplate
 
 
@@ -129,6 +122,15 @@ def content_blocks(self) -> list[types.ContentBlock]:
 
         """
         from langchain_core.messages import content as types
+        from langchain_core.messages.block_translators.anthropic import (
+            _convert_to_v1_from_anthropic_input,
+        )
+        from langchain_core.messages.block_translators.langchain_v0 import (
+            _convert_v0_multimodal_input_to_v1,
+        )
+        from langchain_core.messages.block_translators.openai import (
+            _convert_to_v1_from_chat_completions_input,
+        )
 
         blocks: list[types.ContentBlock] = []
 
@@ -143,26 +145,19 @@ def content_blocks(self) -> list[types.ContentBlock]:
                 blocks.append({"type": "text", "text": item})
             elif isinstance(item, dict):
                 item_type = item.get("type")
-                # Try to convert potential v0 format first
-                converted_block = _convert_legacy_v0_content_block_to_v1(item)
-                if converted_block is not item:  # Conversion happened
-                    blocks.append(cast("types.ContentBlock", converted_block))
-                elif item_type is None or item_type not in types.KNOWN_BLOCK_TYPES:
-                    blocks.append(
-                        cast(
-                            "types.ContentBlock",
-                            {"type": "non_standard", "value": item},
-                        )
-                    )
+                if item_type not in types.KNOWN_BLOCK_TYPES:
+                    blocks.append({"type": "non_standard", "value": item})
                 else:
                     blocks.append(cast("types.ContentBlock", item))
 
         # Subsequent passes: attempt to unpack non-standard blocks
-        blocks = _convert_v0_multimodal_input_to_v1(blocks)
-        # blocks = _convert_to_v1_from_anthropic_input(blocks)
-        # ...
-
-        return _convert_to_v1_from_chat_completions_input(blocks)
+        for parsing_step in [
+            _convert_v0_multimodal_input_to_v1,
+            _convert_to_v1_from_chat_completions_input,
+            _convert_to_v1_from_anthropic_input,
+        ]:
+            blocks = parsing_step(blocks)
+        return blocks
 
     def text(self) -> str:
         """Get the text content of the message.

libs/core/langchain_core/messages/block_translators/__init__.py

@@ -45,37 +45,45 @@ def get_translator(
     return PROVIDER_TRANSLATORS.get(provider)
 
 
-def _auto_register_translators() -> None:
-    """Automatically register all available block translators."""
-    import contextlib
-    import importlib
-    import pkgutil
-    from pathlib import Path
-
-    package_path = Path(__file__).parent
-
-    # Discover all sub-modules
-    for module_info in pkgutil.iter_modules([str(package_path)]):
-        module_name = module_info.name
-
-        # Skip the __init__ module and any private modules
-        if module_name.startswith("_"):
-            continue
-
-        if module_info.ispkg:
-            # For subpackages, discover their submodules
-            subpackage_path = package_path / module_name
-            for submodule_info in pkgutil.iter_modules([str(subpackage_path)]):
-                submodule_name = submodule_info.name
-                if not submodule_name.startswith("_"):
-                    with contextlib.suppress(ImportError, AttributeError):
-                        importlib.import_module(
-                            f".{module_name}.{submodule_name}", package=__name__
-                        )
-        else:
-            # Import top-level translator modules
-            with contextlib.suppress(ImportError, AttributeError):
-                importlib.import_module(f".{module_name}", package=__name__)
-
-
-_auto_register_translators()
+def _register_translators() -> None:
+    """Register all translators in langchain-core.
+
+    A unit test ensures all modules in ``block_translators`` are represented here.
+
+    For translators implemented outside langchain-core, they can be registered by
+    calling ``register_translator`` from within the integration package.
+    """
+    from langchain_core.messages.block_translators.anthropic import (
+        _register_anthropic_translator,
+    )
+    from langchain_core.messages.block_translators.bedrock import (
+        _register_bedrock_translator,
+    )
+    from langchain_core.messages.block_translators.bedrock_converse import (
+        _register_bedrock_converse_translator,
+    )
+    from langchain_core.messages.block_translators.google_genai import (
+        _register_google_genai_translator,
+    )
+    from langchain_core.messages.block_translators.google_vertexai import (
+        _register_google_vertexai_translator,
+    )
+    from langchain_core.messages.block_translators.groq import _register_groq_translator
+    from langchain_core.messages.block_translators.ollama import (
+        _register_ollama_translator,
+    )
+    from langchain_core.messages.block_translators.openai import (
+        _register_openai_translator,
+    )
+
+    _register_bedrock_translator()
+    _register_bedrock_converse_translator()
+    _register_anthropic_translator()
+    _register_google_genai_translator()
+    _register_google_vertexai_translator()
+    _register_groq_translator()
+    _register_ollama_translator()
+    _register_openai_translator()
+
+
+_register_translators()