fix(llms): accept and forward **kwargs in Together/LangChain/Sarvam providers (#5556)

mem0/llms/langchain.py

@@ -57,6 +57,7 @@ class LangchainLLM(LLMBase):
         response_format=None,
         tools: Optional[List[Dict]] = None,
         tool_choice: str = "auto",
+        **kwargs,
     ):
         """
         Generate a response based on the given messages using langchain_community.
@@ -66,6 +67,8 @@ class LangchainLLM(LLMBase):
             response_format (str or object, optional): Format of the response. Not used in Langchain.
             tools (list, optional): List of tools that the model can call.
             tool_choice (str, optional): Tool choice method.
+            **kwargs: Additional model parameters forwarded to the underlying LangChain
+                model's ``invoke`` (matches the ``LLMBase.generate_response`` contract).
 
         Returns:
             str: The generated response.
@@ -90,5 +93,5 @@ class LangchainLLM(LLMBase):
         if tools:
             langchain_model = langchain_model.bind_tools(tools=tools, tool_choice=tool_choice)
 
-        response: AIMessage = langchain_model.invoke(langchain_messages)
+        response: AIMessage = langchain_model.invoke(langchain_messages, **kwargs)
         return self._parse_response(response, tools)

mem0/llms/sarvam.py

@@ -28,7 +28,7 @@ class SarvamLLM(LLMBase):
             getattr(self.config, "sarvam_base_url", None) or os.getenv("SARVAM_API_BASE") or "https://api.sarvam.ai/v1"
         )
 
-    def generate_response(self, messages: List[Dict[str, str]], response_format=None) -> str:
+    def generate_response(self, messages: List[Dict[str, str]], response_format=None, **kwargs) -> str:
         """
         Generate a response based on the given messages using Sarvam-M.
 
@@ -36,6 +36,8 @@ class SarvamLLM(LLMBase):
             messages (list): List of message dicts containing 'role' and 'content'.
             response_format (str or object, optional): Format of the response.
                                                      Currently not used by Sarvam API.
+            **kwargs: Additional provider-specific parameters forwarded to the Sarvam
+                request payload (matches the ``LLMBase.generate_response`` contract).
 
         Returns:
             str: The generated response.
@@ -72,6 +74,9 @@ class SarvamLLM(LLMBase):
                 if param in self.config.model:
                     params[param] = self.config.model[param]
 
+        # Forward any per-call provider-specific parameters (LLMBase contract).
+        params.update(kwargs)
+
         try:
             response = requests.post(url, headers=headers, json=params, timeout=30)
             response.raise_for_status()

mem0/llms/together.py

@@ -58,6 +58,7 @@ class TogetherLLM(LLMBase):
         response_format=None,
         tools: Optional[List[Dict]] = None,
         tool_choice: str = "auto",
+        **kwargs,
     ):
         """
         Generate a response based on the given messages using TogetherAI.
@@ -67,6 +68,8 @@ class TogetherLLM(LLMBase):
             response_format (str or object, optional): Format of the response. Defaults to "text".
             tools (list, optional): List of tools that the model can call. Defaults to None.
             tool_choice (str, optional): Tool choice method. Defaults to "auto".
+            **kwargs: Additional provider-specific parameters forwarded to the Together
+                client (matches the ``LLMBase.generate_response`` contract).
 
         Returns:
             str: The generated response.
@@ -78,6 +81,7 @@ class TogetherLLM(LLMBase):
             "max_tokens": self.config.max_tokens,
             "top_p": self.config.top_p,
         }
+        params.update(kwargs)
         if response_format:
             params["response_format"] = response_format
         if tools:  # TODO: Remove tools if no issues found with new memory addition logic

tests/llms/test_langchain.py

@@ -115,6 +115,20 @@ def test_generate_response_with_tools(mock_langchain_model):
     assert response["tool_calls"][0]["arguments"] == {"data": "Today is a sunny day."}
 
 
+def test_generate_response_forwards_extra_kwargs(mock_langchain_model):
+    """Per the LLMBase contract, extra model kwargs must be accepted and forwarded to
+    the underlying LangChain model's ``invoke``."""
+    config = BaseLlmConfig(model=mock_langchain_model, temperature=0.7, max_tokens=100, api_key="test-api-key")
+    llm = LangchainLLM(config)
+    messages = [{"role": "user", "content": "Hello"}]
+
+    response = llm.generate_response(messages, frequency_penalty=0.5)
+
+    mock_langchain_model.invoke.assert_called_once()
+    assert mock_langchain_model.invoke.call_args.kwargs["frequency_penalty"] == 0.5
+    assert response == "This is a test response"
+
+
 def test_invalid_model():
     """Test that LangchainLLM raises an error with an invalid model."""
     config = BaseLlmConfig(model="not-a-valid-model-instance", temperature=0.7, max_tokens=100, api_key="test-api-key")

tests/llms/test_sarvam.py

@@ -0,0 +1,46 @@
+from unittest.mock import Mock, patch
+
+import pytest
+
+from mem0.configs.llms.base import BaseLlmConfig
+from mem0.llms.sarvam import SarvamLLM
+
+
+@pytest.fixture
+def sarvam_llm():
+    config = BaseLlmConfig(model="sarvam-m", temperature=0.7, max_tokens=100, top_p=1.0, api_key="test-api-key")
+    return SarvamLLM(config)
+
+
+def _mock_post(content="Hello there!"):
+    mock_response = Mock()
+    mock_response.raise_for_status.return_value = None
+    mock_response.json.return_value = {"choices": [{"message": {"content": content}}]}
+    return mock_response
+
+
+def test_generate_response_returns_content(sarvam_llm):
+    with patch("mem0.llms.sarvam.requests.post", return_value=_mock_post("Hi!")) as mock_post:
+        messages = [
+            {"role": "system", "content": "You are a helpful assistant."},
+            {"role": "user", "content": "Hello, how are you?"},
+        ]
+        response = sarvam_llm.generate_response(messages)
+
+    assert response == "Hi!"
+    sent_payload = mock_post.call_args.kwargs["json"]
+    assert sent_payload["model"] == "sarvam-m"
+    assert sent_payload["messages"] == messages
+    assert sent_payload["temperature"] == 0.7
+
+
+def test_generate_response_forwards_extra_kwargs(sarvam_llm):
+    """Per the LLMBase contract, extra provider-specific kwargs must be accepted and
+    forwarded into the Sarvam request payload."""
+    with patch("mem0.llms.sarvam.requests.post", return_value=_mock_post("Hi!")) as mock_post:
+        messages = [{"role": "user", "content": "Hello"}]
+        response = sarvam_llm.generate_response(messages, frequency_penalty=0.5)
+
+    assert response == "Hi!"
+    sent_payload = mock_post.call_args.kwargs["json"]
+    assert sent_payload["frequency_penalty"] == 0.5

tests/llms/test_together.py

@@ -84,3 +84,21 @@ def test_generate_response_with_tools(mock_together_client):
     assert len(response["tool_calls"]) == 1
     assert response["tool_calls"][0]["name"] == "add_memory"
     assert response["tool_calls"][0]["arguments"] == {"data": "Today is a sunny day."}
+
+
+def test_generate_response_forwards_extra_kwargs(mock_together_client):
+    """Per the LLMBase contract, extra provider-specific kwargs must be accepted and
+    forwarded to the Together client (matching openai/deepseek/vllm/xai behavior)."""
+    config = BaseLlmConfig(model="mistralai/Mixtral-8x7B-Instruct-v0.1", temperature=0.7, max_tokens=100, top_p=1.0)
+    llm = TogetherLLM(config)
+    messages = [{"role": "user", "content": "Hello"}]
+
+    mock_response = Mock()
+    mock_response.choices = [Mock(message=Mock(content="Hi"))]
+    mock_together_client.chat.completions.create.return_value = mock_response
+
+    response = llm.generate_response(messages, frequency_penalty=0.5)
+
+    assert response == "Hi"
+    call_kwargs = mock_together_client.chat.completions.create.call_args.kwargs
+    assert call_kwargs["frequency_penalty"] == 0.5