Hi people Is it possible to combine the

At a glance

Hi, people. Is it possible to combine the HuggingFaceEmbeddings service context with the gpt-3.5-turbo model service context? I tried to test an implementation but when loading the index and making the search query, I am getting this error: {"status":{"error":"Wrong input: Vector inserting error: expected dim: 768, got 1536"}

6 comments

LLogan M

Definitely possible! But if you change embeddings you'll need to start with a new index. Also don't forget the service context when loading from disk

Plain Text

llm_predictor = LLMPredictor(llm=ChatOpenAI(temperature=0, model_name='gpt-3.5-turbo'))
embed_model = LangchainEmbedding(HuggingFaceEmbeddings())
service_context = ServiceContext.from_defaults(llm_predictor=llm_predictor, embed_model=embed_model)

index = GPTSimpleVectorIndex.from_documents(documents, service_context=service_context)
...
index.load_from_disk("index.json", service_context=service_context)

eestebanpdl

Thank you, Logan. This is great. Does HuggingFaceEmbeddings() is using all-mpnet-base-v2 by default?

LLogan M

It does! You can also pass In the name of any model from huggingface as well

eestebanpdl

Thanks, Logan!

eestebanpdl

Results change a lot 🤯
Definitely, text-embedding-ada-002 is the best one.

LLogan M

Yea, openAI embeddings are hard to beat. Plus, they are pretty cheap haha

Add a reply

Find answers from the community

Hi people Is it possible to combine the