The community member is having an issue with ChromaDB server embeddings, where the QueryEngine is returning a "Type is not JSON serializable: numpy.float64" error. They found a fix in a pull request for the llama_index.core, but they are using the legacy version due to JinaAI needs, which is not included in the core. Another community member suggests that the Jina embeddings not from the legacy version should use the fix, but there will be no further fixes applied to the legacy version. The community members discuss finding a non-legacy Jina version and the availability of ChromaVectorStore in the core. Eventually, a community member provides a solution involving installing specific llama-index packages and using the HuggingFaceEmbedding and ChromaVectorStore classes, which the original community member finds helpful but difficult to find in the documentation.
I have a little issue with ChromaDB server embeddings, QueryEngine is returning: Type is not JSON serializable: numpy.float64. I found this https://github.com/run-llama/llama_index/pull/11458/files which seems to fix the issue, but it's llama_index.core only, how can I do if I'm using legacy (due to JinaAI needs, not included in core)? thanks a lot
Jina has their own package for using their api. There's also a general huggi face embedding package for using them locally (this might be what you are already using from legacy?)
