Repository navigation
bench(index): refresh streaming memory study on current dedup - #215
Open
adarshh347 wants to merge 2 commits into
Open
adarshh347 wants to merge 2 commits into
adarshh347 wants to merge 2 commits into
Conversation
Add an isolated streaming prototype, real-module regression tests, and reviewed synthetic benchmark reports. Preserve production defaults and legacy dedup semantics; document ANN sampling and production-design limitations. Proposal: StarTrail-org#179
|
@adarshh347 is attempting to deploy a commit to the andylizf's projects Team on Vercel. A member of the Team first needs to authorize it. |
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Related to #179. This is an experimental benchmark and study only; it does not change production behavior, CLI defaults, dependencies, or lockfiles.
The study was originally based on
b4fe763. Current upstream includes #203, which replaced the old fixed-decimal dedup key with range-derived identity and rejects ranges that do not fit in int64. The rebase updates the prototype to use an on-disk SQLite composite key, preserves first-wins ordering, and checks the same range limit. Tests cover the former key-collision pairs and the current range-rejection behavior.Validation:
(384, 16)vector matrix allocation.git diff --checkpassed.At 65,536 × 256 generated rows, median peak RSS changed from 272.73 to 144.52 MiB for IVFFlat and from 264.75 to 76.23 MiB for IVFPQ. These short shared-machine synthetic measurements do not establish production-scale behavior or semantic-retrieval quality. The final IVFFlat index still occupies memory; Linux and representative-data evaluation remain unvalidated.