Skip to content

Support high-dimensional embedding models (3072 and 4096 dim) #49

@chrisolszewski

Description

@chrisolszewski

The semantic index rejects embedding vectors larger than 1024 dimensions (MAX_DIMENSION). Several current models are bigger than that: OpenAI's text-embedding-3-large is 3072, and some local models are 4096. The openai_compatible and ollama backends from #11 let you point AFT at these models, but the index build fails as soon as the vectors come back larger than 1024.

Use case

If you run a larger embedding model through LM Studio, Ollama, or an OpenAI-compatible endpoint, you can't build a semantic index at all right now. Raising the cap to 4096 covers the common models and still keeps a fixed upper bound.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions