[Python] Use LM Studio LLM model as embedding model with LangChain

In LM Studio, I have run a OpenAI server exposing a Gemma 3 model:



Now consider the following code:

from langchain_openai import OpenAIEmbeddings

# Load embeddings
embeddings = OpenAIEmbeddings(
model="gemma-3-27b-it-qat",
openai_api_base="http://127.0.0.1:3003/v1/",
openai_api_key="not-needed",
# We need this to avoid error `'input' field must be a string or an array of strings`
# See: https://github.com/langchain-ai/langchain/issues/21318
check_embedding_ctx_length=False,
)

# Test embeddings
input_text = "The meaning of life is 42"
vector = embeddings.embed_query("hello")
print(vector[:3])


It returns the following error:

---------------------------------------------------------------------------
NotFoundError Traceback (most recent call last)
Cell In[4], line 12
10 # Test embeddings
11 input_text = "The meaning of life is 42"
---> 12 vector = embeddings.embed_query("hello")
13 print(vector[:3])

File c:\Users\USER\miniforge3\envs\langchain\Lib\site-packages\langchain_openai\embeddings\base.py:617, in OpenAIEmbeddings.embed_query(self, text)
608 def embed_query(self, text: str) -> list[float]:
609 """Call out to OpenAI's embedding endpoint for embedding query text.
610
611 Args:
(...) 615 Embedding for the text.
616 """
--> 617 return self.embed_documents([text])[0]

File c:\Users\USER\miniforge3\envs\langchain\Lib\site-packages\langchain_openai\embeddings\base.py:565, in OpenAIEmbeddings.embed_documents(self, texts, chunk_size)
563 embeddings: list[list[float]] = []
564 for i in range(0, len(texts), chunk_size_):
--> 565 response = self.client.create(
566 input=texts[i : i + chunk_size_], **self._invocation_params
567 )
568 if not isinstance(response, dict):
569 response = response.dict()

File c:\Users\USER\miniforge3\envs\langchain\Lib\site-packages\openai\resources\embeddings.py:128, in Embeddings.create(self, input, model, dimensions, encoding_format, user, extra_headers, extra_query, extra_body, timeout)
122 embedding.embedding = np.frombuffer( # type: ignore[no-untyped-call]
123 base64.b64decode(data), dtype="float32"
124 ).tolist()
126 return obj
--> 128 return self._post(
129 "/embeddings",
130 body=maybe_transform(params, embedding_create_params.EmbeddingCreateParams),
131 options=make_request_options(
132 extra_headers=extra_headers,
133 extra_query=extra_query,
134 extra_body=extra_body,
135 timeout=timeout,
136 post_parser=parser,
137 ),
138 cast_to=CreateEmbeddingResponse,
139 )

File c:\Users\USER\miniforge3\envs\langchain\Lib\site-packages\openai\_base_client.py:1276, in SyncAPIClient.post(self, path, cast_to, body, options, files, stream, stream_cls)
1262 def post(
1263 self,
1264 path: str,
(...) 1271 stream_cls: type[_StreamT] | None = None,
1272 ) -> ResponseT | _StreamT:
1273 opts = FinalRequestOptions.construct(
1274 method="post", url=path, json_data=body, files=to_httpx_files(files), **options
1275 )
-> 1276 return cast(ResponseT, self.request(cast_to, opts, stream=stream, stream_cls=stream_cls))

File c:\Users\USER\miniforge3\envs\langchain\Lib\site-packages\openai\_base_client.py:949, in SyncAPIClient.request(self, cast_to, options, remaining_retries, stream, stream_cls)
946 else:
947 retries_taken = 0
--> 949 return self._request(
950 cast_to=cast_to,
951 options=options,
952 stream=stream,
953 stream_cls=stream_cls,
954 retries_taken=retries_taken,
955 )

File c:\Users\USER\miniforge3\envs\langchain\Lib\site-packages\openai\_base_client.py:1057, in SyncAPIClient._request(self, cast_to, options, retries_taken, stream, stream_cls)
1054 err.response.read()
1056 log.debug("Re-raising status error")
-> 1057 raise self._make_status_error_from_response(err.response) from None
1059 return self._process_response(
1060 cast_to=cast_to,
1061 options=options,
(...) 1065 retries_taken=retries_taken,
1066 )

NotFoundError: Error code: 404 - {'error': {'message': 'No models loaded. Please load a model in the developer page or use the `lms load` command.', 'type': 'invalid_request_error', 'param': 'model', 'code': 'model_not_found'}}


However, if I load an embedding model, for instance:



Then it works and the code returns:

[-0.0032659093849360943, 0.017258411273360252, -0.015825312584638596]


Notice that this works even if model="gemma-3-27b-it-qat" stays as it is (and the embedding model I loaded isn't that).

I thought I should/could use the same LLM for the embedding.

How can I use the same LLM for the embeddings?

If I can't, what embedding model should I use?

Continue reading...