multiple: permit optional fields on multimodal content blocks (#30887)

Instead of stuffing provider-specific fields in `metadata`, they can go
directly on the content block.

libs/core/langchain_core/messages/content_blocks.py

@@ -6,13 +6,11 @@ from pydantic import TypeAdapter, ValidationError
 from typing_extensions import NotRequired, TypedDict
 
 
-class BaseDataContentBlock(TypedDict):
+class BaseDataContentBlock(TypedDict, total=False):
     """Base class for data content blocks."""
 
     mime_type: NotRequired[str]
     """MIME type of the content block (if needed)."""
-    metadata: NotRequired[dict]
-    """Provider-specific metadata such as citations or filenames."""
 
 
 class URLContentBlock(BaseDataContentBlock):

libs/core/tests/unit_tests/test_messages.py

@@ -1107,6 +1107,15 @@ def test_is_data_content_block() -> None:
             "mime_type": "image/jpeg",
         }
     )
+    assert is_data_content_block(
+        {
+            "type": "image",
+            "source_type": "base64",
+            "data": "<base64 data>",
+            "mime_type": "image/jpeg",
+            "cache_control": {"type": "ephemeral"},
+        }
+    )
     assert is_data_content_block(
         {
             "type": "image",
@@ -1148,7 +1157,7 @@ def test_convert_to_openai_image_block() -> None:
         "type": "image",
         "source_type": "url",
         "url": "https://...",
-        "metadata": {"cache_control": {"type": "ephemeral"}},
+        "cache_control": {"type": "ephemeral"},
     }
     expected = {
         "type": "image_url",
@@ -1162,7 +1171,7 @@ def test_convert_to_openai_image_block() -> None:
         "source_type": "base64",
         "data": "<base64 data>",
         "mime_type": "image/jpeg",
-        "metadata": {"cache_control": {"type": "ephemeral"}},
+        "cache_control": {"type": "ephemeral"},
     }
     expected = {
         "type": "image_url",

libs/partners/anthropic/langchain_anthropic/chat_models.py

@@ -239,11 +239,12 @@ def _format_data_content_block(block: dict) -> dict:
     else:
         raise ValueError(f"Block of type {block['type']} is not supported.")
 
-    if formatted_block and (metadata := block.get("metadata")):
-        if "cache_control" in metadata:
-            formatted_block["cache_control"] = metadata["cache_control"]
-        if "citations" in metadata:
-            formatted_block["citations"] = metadata["citations"]
+    if formatted_block:
+        for key in ["cache_control", "citations", "title", "context"]:
+            if key in block:
+                formatted_block[key] = block[key]
+            elif (metadata := block.get("metadata")) and key in metadata:
+                formatted_block[key] = metadata[key]
 
     return formatted_block
 

libs/partners/anthropic/tests/unit_tests/test_chat_models.py

@@ -703,7 +703,7 @@ def test__format_messages_with_cache_control() -> None:
                     "source_type": "base64",
                     "mime_type": "application/pdf",
                     "data": "<base64 data>",
-                    "metadata": {"cache_control": {"type": "ephemeral"}},
+                    "cache_control": {"type": "ephemeral"},
                 },
             ]
         )
@@ -742,7 +742,7 @@ def test__format_messages_with_citations() -> None:
                     "source_type": "text",
                     "text": "The grass is green. The sky is blue.",
                     "mime_type": "text/plain",
-                    "metadata": {"citations": {"enabled": True}},
+                    "citations": {"enabled": True},
                 },
                 {"type": "text", "text": "What color is the grass and sky?"},
             ]

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

@@ -194,14 +194,16 @@ def _format_data_content_block(block: dict) -> dict:
     elif block["type"] == "file":
         if block["source_type"] == "base64":
             file = {"file_data": f"data:{block['mime_type']};base64,{block['data']}"}
-            if (metadata := block.get("metadata")) and ("filename" in metadata):
+            if filename := block.get("filename"):
+                file["filename"] = filename
+            elif (metadata := block.get("metadata")) and ("filename" in metadata):
                 file["filename"] = metadata["filename"]
             else:
                 warnings.warn(
                     "OpenAI may require a filename for file inputs. Specify a filename "
-                    "in the metadata: {'type': 'file', 'source_type': 'base64', "
+                    "in the content block: {'type': 'file', 'source_type': 'base64', "
                     "'mime_type': 'application/pdf', 'data': '...', "
-                    "'metadata': {'filename': 'my-pdf'}}"
+                    "'filename': 'my-pdf'}"
                 )
             formatted_block = {"type": "file", "file": file}
         elif block["source_type"] == "id":

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

@@ -96,7 +96,7 @@ class TestOpenAIStandard(ChatModelIntegrationTests):
                     "source_type": "base64",
                     "mime_type": "application/pdf",
                     "data": pdf_data,
-                    "metadata": {"filename": "my-pdf"},  # OpenAI requires a filename
+                    "filename": "my-pdf",  # OpenAI requires a filename
                 },
             ]
         )

libs/partners/openai/tests/unit_tests/chat_models/test_base.py

@@ -676,7 +676,7 @@ def test_format_message_content() -> None:
             "source_type": "base64",
             "data": "<base64 data>",
             "mime_type": "application/pdf",
-            "metadata": {"filename": "my_file"},
+            "filename": "my_file",
         }
     ]
     expected = [