vLLM issue for embedding models

[corradodebari]

Checklist

• I have searched the existing issues for similar issues.
• I added a very descriptive title to this issue.
• I have provided sufficient information below to help reproduce this issue.

Summary

Adding a new embedding models:
modelID: nomic-ai/nomic-embed-text-v1
Provider: openai_compatible
Provider URL: http://gpu:8001/v1
Max Chunk Size: 8192
that can be reached successfully with:

curl http://gpu:8001/v1/embeddings \
  -H "Content-Type: application/json" \
  -d '{
    "model": "nomic-ai/nomic-embed-text-v1",
    "input": ["hello world", "this is an embedding test"]
  }'

getting:
{"id":"embd-47709d027fbe49c6b6ddda219c98ff66","object":"list","created":1756729440,"model":"nomic-ai/nomic-embed-text-v1","data":[{"index":0,"object":"embedding","embedding":[0.016637714579701424,...

Using for Split/Embed Tool:

2025-Sep-01 14:25:42 (v1.1.5.dev37+g039d0f4be.d20250827) - ERROR    - (langchain_community.vectorstores.oraclevs): An unexpected error occurred while creating the index.
Traceback (most recent call last):
  File "/Users/cdebari/Documents/GitHub/ai-optimizer-248-mcp-export/src/.venv/lib/python3.11/site-packages/langchain_community/vectorstores/oraclevs.py", line 529, in __init__
    embedding_dim = self.get_embedding_dimension()
                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/cdebari/Documents/GitHub/ai-optimizer-248-mcp-export/src/.venv/lib/python3.11/site-packages/langchain_community/vectorstores/oraclevs.py", line 569, in get_embedding_dimension
    embedded_document = self._embed_documents(
                        ^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/cdebari/Documents/GitHub/ai-optimizer-248-mcp-export/src/.venv/lib/python3.11/site-packages/langchain_community/vectorstores/oraclevs.py", line 578, in _embed_documents
    return self.embedding_function.embed_documents(texts)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/cdebari/Documents/GitHub/ai-optimizer-248-mcp-export/src/.venv/lib/python3.11/site-packages/langchain_openai/embeddings/base.py", line 591, in embed_documents
    return self._get_len_safe_embeddings(
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/cdebari/Documents/GitHub/ai-optimizer-248-mcp-export/src/.venv/lib/python3.11/site-packages/langchain_openai/embeddings/base.py", line 479, in _get_len_safe_embeddings
    response = self.client.create(
               ^^^^^^^^^^^^^^^^^^^
  File "/Users/cdebari/Documents/GitHub/ai-optimizer-248-mcp-export/src/.venv/lib/python3.11/site-packages/openai/resources/embeddings.py", line 132, in create
    return self._post(
           ^^^^^^^^^^^
  File "/Users/cdebari/Documents/GitHub/ai-optimizer-248-mcp-export/src/.venv/lib/python3.11/site-packages/openai/_base_client.py", line 1259, in post
    return cast(ResponseT, self.request(cast_to, opts, stream=stream, stream_cls=stream_cls))
                           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/cdebari/Documents/GitHub/ai-optimizer-248-mcp-export/src/.venv/lib/python3.11/site-packages/openai/_base_client.py", line 1047, in request
    raise self._make_status_error_from_response(err.response) from None
openai.BadRequestError: Error code: 400 - {'object': 'error', 'message': 'Token id 60720 is out of vocabulary', 'type': 'BadRequestError', 'param': None, 'code': 400}
2025-Sep-01 14:25:42 (v1.1.5.dev37+g039d0f4be.d20250827) - INFO     - (uvicorn.access): 127.0.0.1:57876 - "POST /v1/embed/?client=dab82212-fe3a-4de0-a827-28ab16eb7038 HTTP/1.1" 500

Steps To Reproduce

No response

Expected Behavior

No response

Current Behavior

No response

Is this a regression?

• Yes, this used to work in a previous version.

Debug info

• Version:
• Python version:
• Operating System:
• Browser:

Additional Information

No response

[corradodebari]

The langchain_openai.OpenAIEmbeddings has the issue:

#Doesn't work
from langchain_openai import OpenAIEmbeddings

emb = OpenAIEmbeddings(
    model="intfloat/e5-mistral-7b-instruct",
    openai_api_base="http://gpu:8001/v1",
    openai_api_key="EMPTY",
)

# E5-style prefix is recommended (not required for the OOV fix)
text = "query: LangChain is the framework for building context-aware reasoning applications"
vec = emb.embed_query(text)
print(len(vec), vec[:5])

direct OpenAI Embeddings works:

#It works
from openai import OpenAI

client = OpenAI(base_url="http://gpu:8001/v1", api_key="EMPTY")

# minimal probe identical to your curl
r = client.embeddings.create(model="intfloat/e5-mistral-7b-instruct", input="hello")
print(len(r.data[0].embedding))

# now your real query (E5 likes "query:" prefix)
q = "query: LangChain is the framework for building context-aware reasoning applications"
r = client.embeddings.create(model="intfloat/e5-mistral-7b-instruct",
                             input=q,
                             encoding_format="float")
print(len(r.data[0].embedding), r.data[0].embedding[:5])

[corradodebari]

The OpenAIEmbeddings works disabling the OpenAI tokenizer, tiktoken, with:

from langchain_openai import OpenAIEmbeddings

emb = OpenAIEmbeddings(
    model="intfloat/e5-mistral-7b-instruct",
    openai_api_base="http://gpu:8001/v1",
    openai_api_key="EMPTY",
     check_embedding_ctx_length=False      #TO DISABLE tiktoken
)