privateGPT/private_gpt/components/embedd...

Hi everyone

I'm trying to use PrivateGPT which is mostly based on llama_index. For now I have an issue when running embeding inference with on aws sagemaker. I'm using the BAAI/bge-large-en-v1.5 embedding which, I think, is quite common.

In PGPT, sagemaker embedding is made through Custom embedding in llama_index, the current implementation is available here

For now :

• My model is deployed

• PGPT is able to request the model
• The model send back a vector
• (I think) that this vector do not fit my vector database possibly for a wrong dimensionality

Currently I have the following issue (in attachment).

Do you guys have any idea where to look ?

For now my guess is to look for some parameters like max_length (that would reduce the dimension of my vector) but don't know if it's a good idea and where to look ... Any help appreciated 🙂