Matt
Matt
As a user, I want to upload binary embeddings from cohere (et al) directly into the database. This helps support the latest and greatest embedding formats, run faster, and use...
Google released XTR that uses late interaction. https://github.com/google-deepmind/xtr We should be able to support this and enable users to run this themselves.
In order to enable filtering of documents, we should be able to formulate query trees. ## What is a Query Tree A query tree should be an executable graph on...
As a user, I want to understand why certain documents are returned so that we can formulate the LLM context better. One of the benefits of late interaction is having...
As a user of LintDB, I want to be able to filter by metadata in a search so that I can more accurately retrieve my data. ### Acceptance Criteria 1....
My system is currently under pressure while training a new model. Benchmarking the search critical path, it's the coarse quantizer that takes the most time at a whopping 80%. ...
We use sgemm to run matmuls for every query. We can cut this in half by supporting float16, which MKL/OpenBLAS support. Requirements: - We need to be able to toggle...
resolves #45 FieldValues now accept float16.
Since late interaction is already so expensive storage-wise, we can support float16 as a quick win to reduce storage costs. This would involve adding support for a new datatype. Alternatively,...
As with all other vector databases, we should support traditional search as well. While sparse search isn't as necessary with colbert-style retrieval, it gives us more options to tradeoff latency...