lance-format / lance-format/lance
Vector search performance optimization roadmap
Open
@BubbleCal is already working on this.
Since May 18, 2026.
A-index
enhancement
epic
performance
- Dominant language
- Rust
- Stars
- 7.1k
- Forks
- 852
- Avg merge
- 3d 18h
- Merged PRs (30d)
- 272
Description
Summary
Track a focused sequence of vector query performance work in Lance.
The main themes are:
- Batch vector query support: one query request can contain multiple query vectors.
- Flat KNN MVP: share table scan/load work across the batch and return per-query top-k results.
- Fused batch distance kernels: avoid turning batch queries into a loop over single-query kernels.
- Metric-specific kernels: especially small-dimensional Hamming-like metrics where SIMD utilization can matter.
- Pruning: partition-level pruning now that vector search can process partitions serially, plus in-partition pruning such as PDX-style approaches.
- Benchmarks and guardrails: measure p50/p95/p99 latency, QPS, recall, and regression risk across dimensions, metrics, and partition sizes.
Linear project: Vector Search Performance Optimizations
Linear parent: lance-format/lance#6820
Tasks
- lance-format/lance#6821 / lance-format/lance#6821
- lance-format/lance#6822 / lance-format/lance#6822
- lance-format/lance#6823 / lance-format/lance#6823
- lance-format/lance#6824 / lance-format/lance#6824
- lance-format/lance#6825 / lance-format/lance#6825
- lance-format/lance#6826 / lance-format/lance#6826
- lance-format/lance#6827 / lance-format/lance#6827
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Assessment
This issue has not been assessed yet.