Qwen

/

Qwen3-Embedding-8B

like 786

Follow

Qwen 102k

I achieved good results using the gte-Qwen2-7B-instruct model. However, when switching to Qwen/Qwen3-Embedding-8B via Ollama with the same code, the performance dropped significantly. Could there be something wrong with how I’m using it?
Code here:

from langchain_ollama import OllamaEmbeddings
from langchain_chroma import Chroma

def vector_index(chunks: list):
    Chroma.from_documents(
        documents = chunks,
        embedding = OllamaEmbeddings(model=config.VEC_MODEL),
        persist_directory = config.VEC_DB_DIR,
    )

def test():
    query_str = 'some questions'
    vector_store = Chroma(persist_directory=config.VEC_DB_DIR, 
                          embedding_function=OllamaEmbeddings(model=config.VEC_MODEL) )
    retriever = vector_store.as_retriever(search_type='mmr', search_kwargs={"k": 20, "score_threshold": 0.2} )
    retrieved_docs = retriever.invoke(query_str, k=4)