Daniel Chalef
Daniel Chalef
`faiss-cpu` indexes are CPU arch-dependent. You can't build an index on `arm64` (a MacBook M-series) and deploy to a `amd64` server (most Intel and AMD-powered cloud services). As pointed out...
I no longer use supbase and, unfortunately, don't have time to dig in to optimize this. The VectorStore can likely be simplified and improved significantly by refactoring to use the...
@winglian Yes, I've tried bitsandbytes versions: - main/3ba076d3fb3f1590a049f5715c446514cff3a8c0 - 0.42.0 - 0.41.0 - and whatever version were used in recent builds of `winglian/axolotl:main-py3.10-*`
The tensorrt-llm backend requires setting the `gpt_model_path`. This can't be relative and fails with S3-based model repos. Any update on this @kthui?
@ogabrielluiz I solved the issue with rendering of field names and the drop-down. I needed to rebuild the front-end. I've also got the ZepMemory component working. The session_id is generated...
Hey @Elsayed91 , I've been heads down working on some other projects. Hope to get back to this soon!
Hey @QuinnGT, thanks for the suggestion. We're refactoring how LLM inference works in Zep. We'll consider how Bedrock might be supported. We already support Anthropic models, albeit directly via the...
Unfortunately, we don't have the bandwidth to debug the Kubernetes deployment. We would most certainly appreciate a contribution if you happen to fix this.
Zep stores both embeddings and the document contents, so you may be able to use Zep's retriever for both of these.
Fixed in #287 Will be available in the next release of Zep. Likely today.