Closed kristiankielhofner closed 8 months ago
Weaviate has many features missing for h2oGPT, so I'm happy if someone is able to fix this problem but I won't focus on it for that reason. Thanks!
No worries, I went a different direction. Thanks for all of your work on the project!
Fantastic project! Working perfectly for my use case with one exception. I can't seem to get h2ogpt to work with a Weaviate server.
h2ogpt version - main@338fa54
Start Weaviate:
Make DB:
I can confirm doc chunks are added to Weaviate. For example:
curl 'http://localhost:8080/v1/objects/035b93dc-b3bd-452f-8cb1-934047bbcecf?include=vector' | jq
Start h2ogpt:
In h2ogpt UI:
Response:
h2ogpt console output:
I've tried every possible combination of Weaviate server versions, various h2ogpt releases, different models, load model directly/vLLM/HF TGI, different sentence embedding models, etc but I just can't seem to get this to work. I also tried with embedded Weaviate and verified
/workspace/db_dir_UserData
is populated correctly. Same result fromgen.py
.Chroma works perfectly when set with
db_type=chroma
.What am I missing?