core[patch]: Allow bound models as token_counter in trim_messages (#25563)

libs/core/langchain_core/messages/utils.py

@@ -514,6 +514,8 @@ def merge_message_runs(
     return merged
 
 
+# TODO: Update so validation errors (for token_counter, for example) are raised on
+# init not at runtime.
 @_runnable_support
 def trim_messages(
     messages: Union[Iterable[MessageLikeRepresentation], PromptValue],
@@ -759,24 +761,30 @@ def trim_messages(
                     AIMessage("This is a 4 token text. The full message is 10 tokens.", id="fourth"),
                 ]
     """  # noqa: E501
-    from langchain_core.language_models import BaseLanguageModel
 
     if start_on and strategy == "first":
         raise ValueError
     if include_system and strategy == "first":
         raise ValueError
     messages = convert_to_messages(messages)
-    if isinstance(token_counter, BaseLanguageModel):
+    if hasattr(token_counter, "get_num_tokens_from_messages"):
-        list_token_counter = token_counter.get_num_tokens_from_messages
+        list_token_counter = getattr(token_counter, "get_num_tokens_from_messages")
-    elif (
+    elif callable(token_counter):
-        list(inspect.signature(token_counter).parameters.values())[0].annotation
+        if (
-        is BaseMessage
+            list(inspect.signature(token_counter).parameters.values())[0].annotation
-    ):
+            is BaseMessage
+        ):
 
-        def list_token_counter(messages: Sequence[BaseMessage]) -> int:
+            def list_token_counter(messages: Sequence[BaseMessage]) -> int:
-            return sum(token_counter(msg) for msg in messages)  # type: ignore[arg-type, misc]
+                return sum(token_counter(msg) for msg in messages)  # type: ignore[arg-type, misc]
+        else:
+            list_token_counter = token_counter  # type: ignore[assignment]
     else:
-        list_token_counter = token_counter  # type: ignore[assignment]
+        raise ValueError(
+            f"'token_counter' expected ot be a model that implements "
+            f"'get_num_tokens_from_messages()' or a function. Received object of type "
+            f"{type(token_counter)}."
+        )
 
     try:
         from langchain_text_splitters import TextSplitter

libs/core/tests/unit_tests/messages/test_utils.py

@@ -2,6 +2,7 @@ from typing import Dict, List, Type
 
 import pytest
 
+from langchain_core.language_models.fake_chat_models import FakeChatModel
 from langchain_core.messages import (
     AIMessage,
     BaseMessage,
@@ -316,6 +317,19 @@ def test_trim_messages_invoke() -> None:
     assert actual == expected
 
 
+def test_trim_messages_bound_model_token_counter() -> None:
+    trimmer = trim_messages(
+        max_tokens=10, token_counter=FakeTokenCountingModel().bind(foo="bar")
+    )
+    trimmer.invoke([HumanMessage("foobar")])
+
+
+def test_trim_messages_bad_token_counter() -> None:
+    trimmer = trim_messages(max_tokens=10, token_counter={})
+    with pytest.raises(ValueError):
+        trimmer.invoke([HumanMessage("foobar")])
+
+
 def dummy_token_counter(messages: List[BaseMessage]) -> int:
     # treat each message like it adds 3 default tokens at the beginning
     # of the message and at the end of the message. 3 + 4 + 3 = 10 tokens
@@ -338,3 +352,8 @@ def dummy_token_counter(messages: List[BaseMessage]) -> int:
                 + default_msg_suffix_len
             )
     return count
+
+
+class FakeTokenCountingModel(FakeChatModel):
+    def get_num_tokens_from_messages(self, messages: List[BaseMessage]) -> int:
+        return dummy_token_counter(messages)