Embeddings

Changed it to this and same error

Plain Text

embed_model = LangchainEmbedding(HuggingFaceEmbeddings())
llm_predictor = LLMPredictor(llm=ChatOpenAI(temperature=0.7, model_name="text-curie-001"))
service_context = ServiceContext.from_defaults(embed_model=embed_model, llm_predictor=llm_predictor)
index = GPTSimpleVectorIndex.from_documents(documents, service_context=service_context)
index.save_to_disk('index.json')
index = GPTSimpleVectorIndex.load_from_disk('index.json')
response = index.query("How do i get kudos ?", similarity_top_k=5, mode='embedding', service_context=service_context, optimizer=SentenceEmbeddingOptimizer(percentile_cutoff=0.5))

᲼

11 comments

LLogan M

Does it work if you remove the percentile cutoff? Or passing in the service_context when loading from disk too? (Also no need to set the mode for vector indexes)

᲼᲼᲼᲼᲼᲼᲼᲼᲼᲼᲼᲼

Neither did work :/

᲼᲼᲼᲼᲼᲼᲼᲼᲼᲼᲼᲼

Maybe it's due to using curie ?

LLogan M

Well, you should be able to switch embedding models whenever... but maybe curie isn't made for embeddings lol I never looked into it

᲼᲼᲼᲼᲼᲼᲼᲼᲼᲼᲼᲼

davinci doesnt work either xD

LLogan M

I think openai only advertises one model for embeddings 👀

LLogan M

Weird

yyoelk

Yeah they used to have several but I think they had consolidated them into one - text-embedding-ada-002
@᲼᲼᲼᲼᲼᲼᲼᲼᲼᲼᲼ you might wanna try this one

᲼᲼᲼᲼᲼᲼᲼᲼᲼᲼᲼᲼

tbh i was wanting to use a local one

LLogan M

You can load any model locally from huggingface using this: https://gpt-index.readthedocs.io/en/latest/how_to/customization/embeddings.html#custom-embeddings

By default it's loading https://huggingface.co/sentence-transformers/all-mpnet-base-v2 but you can pass in any model name from huggingface

᲼᲼᲼᲼᲼᲼᲼᲼᲼᲼᲼᲼

yeah that's what i did but this + optimizer doesnt work, gonna look into using a local Alpaca or Llama to reduce costs

Add a reply

Find answers from the community

Embeddings