lance-format / lance-format/lance
Train PQ without coping data to build sub-vectors arrays.
Open
Nobody has claimed this yet.
A-index
benchmark
- Dominant language
- Rust
- Stars
- 7.1k
- Forks
- 852
- Avg merge
- 3d 18h
- Merged PRs (30d)
- 272
Description
Problem Statement
For PQ training, right now we copy data to build sub-vectors continuously. flamegraph shows that it contributed more than half of time.
We should figure out a way to read sub-vectors without data copying, for example, make MatrixView accepts strides.
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.
Research direction
Start with the PQ training path and the MatrixView type mentioned in the issue, using the referenced flamegraph to identify where sub-vector arrays are copied. The work is done when PQ training can read sub-vectors without repeatedly copying the data and the profiling result shows the copying cost has been removed or substantially reduced.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- rust
- Domain
- machine-learning, performance
- Issue type
- Refactor
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 35/100