capture response context

libs/partners/ollama/langchain_ollama/_compat.py

@@ -230,6 +230,8 @@ def _convert_to_v1_from_ollama_format(response: dict[str, Any]) -> AIMessageV1:
         metadata_as_dict["eval_count"] = response["eval_count"]
     if "eval_duration" in response:
         metadata_as_dict["eval_duration"] = response["eval_duration"]
+    if "context" in response:
+        metadata_as_dict["context"] = response["context"]
 
     return AIMessageV1(
         content=content,
@@ -296,6 +298,8 @@ def _convert_chunk_to_v1(chunk: dict[str, Any]) -> AIMessageChunkV1:
             response_metadata["eval_count"] = chunk["eval_count"]  # type: ignore[typeddict-unknown-key]
         if "eval_duration" in chunk:
             response_metadata["eval_duration"] = chunk["eval_duration"]  # type: ignore[typeddict-unknown-key]
+        if "context" in chunk:
+            response_metadata["context"] = chunk["context"]  # type: ignore[typeddict-unknown-key]
 
     usage_metadata = None
     if chunk.get("done") is True:

libs/partners/ollama/tests/unit_tests/test_chat_models_v1.py

@@ -124,6 +124,34 @@ class TestMessageConversion:
         assert result.response_metadata.get("model_name") == MODEL_NAME
         assert result.response_metadata.get("done") is True
 
+    def test_convert_from_ollama_format_with_context(self) -> None:
+        """Test converting Ollama response with context field to `AIMessageV1`."""
+        test_context = [1, 2, 3, 4, 5]  # Example tokenized context
+        ollama_response = {
+            "model": MODEL_NAME,
+            "created_at": "2024-01-01T00:00:00Z",
+            "message": {
+                "role": "assistant",
+                "content": "Hello! How can I help you today?",
+            },
+            "done": True,
+            "done_reason": "stop",
+            "total_duration": 1000000,
+            "prompt_eval_count": 10,
+            "eval_count": 20,
+            "context": test_context,
+        }
+
+        result = _convert_to_v1_from_ollama_format(ollama_response)
+
+        assert isinstance(result, AIMessageV1)
+        assert len(result.content) == 1
+        assert result.content[0].get("type") == "text"
+        assert result.content[0].get("text") == "Hello! How can I help you today?"
+        assert result.response_metadata.get("model_name") == MODEL_NAME
+        assert result.response_metadata.get("done") is True
+        assert result.response_metadata.get("context") == test_context
+
     def test_convert_chunk_to_v1(self) -> None:
         """Test converting Ollama streaming chunk to `AIMessageChunkV1`."""
         chunk = {
@@ -139,6 +167,27 @@ class TestMessageConversion:
         assert result.content[0].get("type") == "text"
         assert result.content[0].get("text") == "Hello"
 
+    def test_convert_chunk_to_v1_with_context(self) -> None:
+        """Test converting Ollama streaming chunk with context to `AIMessageChunkV1`."""
+        test_context = [10, 20, 30, 40, 50]  # Example tokenized context
+        chunk = {
+            "model": MODEL_NAME,
+            "created_at": "2024-01-01T00:00:00Z",
+            "message": {"role": "assistant", "content": "Hello"},
+            "done": True,
+            "done_reason": "stop",
+            "context": test_context,
+            "prompt_eval_count": 5,
+            "eval_count": 3,
+        }
+
+        result = _convert_chunk_to_v1(chunk)
+
+        assert len(result.content) == 1
+        assert result.content[0].get("type") == "text"
+        assert result.content[0].get("text") == "Hello"
+        assert result.response_metadata.get("context") == test_context
+
     def test_convert_empty_content(self) -> None:
         """Test converting empty content blocks."""
         message = HumanMessageV1(content=[])
@@ -193,7 +242,12 @@ class TestChatOllamaV1(ChatModelV1UnitTests):
 
     @property
     def supports_non_standard_blocks(self) -> bool:
-        """Override to indicate Ollama doesn't support non-standard content blocks."""
+        """Override to indicate Ollama doesn't support non-standard content blocks.
+
+        So far, everything returned by Ollama fits into the standard
+        `text`, `image`, and `thinking` content blocks.
+
+        """
         return False
 
     @pytest.fixture