cohere[patch]: Add additional kwargs support for Cohere SDK params (#19533)

* Adds support for `additional_kwargs` in `get_cohere_chat_request`
* This functionality passes in Cohere SDK specific parameters from
`BaseMessage` based classes to the API

---------

Co-authored-by: Erick Friis <erick@langchain.dev>

libs/partners/cohere/langchain_cohere/chat_models.py

@@ -47,6 +47,7 @@ def get_role(message: BaseMessage) -> str:
 def get_cohere_chat_request(
     messages: List[BaseMessage],
     *,
+    documents: Optional[List[Dict[str, str]]] = None,
     connectors: Optional[List[Dict[str, str]]] = None,
     **kwargs: Any,
 ) -> Dict[str, Any]:
@@ -60,24 +61,33 @@ def get_cohere_chat_request(
     Returns:
         The request for the Cohere chat API.
     """
-    documents = (
-        None
-        if "source_documents" not in kwargs
-        else [
-            {
-                "snippet": doc.page_content,
-                "id": doc.metadata.get("id") or f"doc-{str(i)}",
-            }
-            for i, doc in enumerate(kwargs["source_documents"])
-        ]
-    )
-    kwargs.pop("source_documents", None)
-    maybe_connectors = connectors if documents is None else None
+    additional_kwargs = messages[-1].additional_kwargs
+
+    # cohere SDK will fail loudly if both connectors and documents are provided
+    if (
+        len(additional_kwargs.get("documents", [])) > 0
+        and documents
+        and len(documents) > 0
+    ):
+        raise ValueError(
+            "Received documents both as a keyword argument and as an prompt additional"
+            "keywword argument. Please choose only one option."
+        )
+
+    formatted_docs = [
+        {
+            "text": doc.page_content,
+            "id": doc.metadata.get("id") or f"doc-{str(i)}",
+        }
+        for i, doc in enumerate(additional_kwargs.get("documents", []))
+    ] or documents
+    if not formatted_docs:
+        formatted_docs = None
 
     # by enabling automatic prompt truncation, the probability of request failure is
     # reduced with minimal impact on response quality
     prompt_truncation = (
-        "AUTO" if documents is not None or connectors is not None else None
+        "AUTO" if formatted_docs is not None or connectors is not None else None
     )
 
     req = {
@@ -85,8 +95,8 @@ def get_cohere_chat_request(
         "chat_history": [
             {"role": get_role(x), "message": x.content} for x in messages[:-1]
         ],
-        "documents": documents,
-        "connectors": maybe_connectors,
+        "documents": formatted_docs,
+        "connectors": connectors,
         "prompt_truncation": prompt_truncation,
         **kwargs,
     }

libs/partners/cohere/tests/integration_tests/test_chat_models.py

@@ -3,7 +3,7 @@ from langchain_cohere import ChatCohere
 
 
 def test_stream() -> None:
-    """Test streaming tokens from OpenAI."""
+    """Test streaming tokens from ChatCohere."""
     llm = ChatCohere()
 
     for token in llm.stream("I'm Pickle Rick"):
@@ -11,7 +11,7 @@ def test_stream() -> None:
 
 
 async def test_astream() -> None:
-    """Test streaming tokens from OpenAI."""
+    """Test streaming tokens from ChatCohere."""
     llm = ChatCohere()
 
     async for token in llm.astream("I'm Pickle Rick"):

libs/partners/cohere/tests/integration_tests/test_rag.py

@@ -0,0 +1,63 @@
+"""Test ChatCohere chat model."""
+
+from typing import Any, Dict, List
+
+from langchain_core.documents import Document
+from langchain_core.messages.human import HumanMessage
+from langchain_core.prompts.chat import ChatPromptTemplate, MessagesPlaceholder
+from langchain_core.runnables import (
+    RunnablePassthrough,
+    RunnableSerializable,
+)
+
+from langchain_cohere import ChatCohere
+
+
+def test_connectors() -> None:
+    """Test connectors parameter support from ChatCohere."""
+    llm = ChatCohere().bind(connectors=[{"id": "web-search"}])
+
+    result = llm.invoke("Who directed dune two? reply with just the name.")
+    assert isinstance(result.content, str)
+
+
+def test_documents() -> None:
+    """Test documents paraneter support from ChatCohere."""
+    docs = [{"text": "The sky is green."}]
+    llm = ChatCohere().bind(documents=docs)
+    prompt = "What color is the sky?"
+
+    result = llm.invoke(prompt)
+    assert isinstance(result.content, str)
+    assert len(result.response_metadata["documents"]) == 1
+
+
+def test_documents_chain() -> None:
+    """Test documents paraneter support from ChatCohere."""
+    llm = ChatCohere()
+
+    def get_documents(_: Any) -> List[Document]:
+        return [Document(page_content="The sky is green.")]
+
+    def format_input_msgs(input: Dict[str, Any]) -> List[HumanMessage]:
+        return [
+            HumanMessage(
+                content=input["message"],
+                additional_kwargs={
+                    "documents": input.get("documents", None),
+                },
+            )
+        ]
+
+    prompt = ChatPromptTemplate.from_messages([MessagesPlaceholder("input_msgs")])
+    chain: RunnableSerializable[Any, Any] = (
+        {"message": RunnablePassthrough(), "documents": get_documents}
+        | RunnablePassthrough()
+        | {"input_msgs": format_input_msgs}
+        | prompt
+        | llm
+    )
+
+    result = chain.invoke("What color is the sky?")
+    assert isinstance(result.content, str)
+    assert len(result.response_metadata["documents"]) == 1