Repository navigation
Conversation
|
All contributors have signed the CLA ✍️ ✅ |
ze33m
force-pushed
the
vchordrq-emeans
branch
from
November 12, 2026 10:18
bc5c9d7 to
0784fbe
Compare
Author
|
I have read the CLA Document and I hereby sign the CLA |
gaocegege
added a commit
to tensorchord/CLA
that referenced
this pull request
Nov 12, 2026
5 of 15 tasks
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
Add opt-in e-means training for leaf centroids in internal
vchordrqindex builds. Enable it with
kmeans_algorithm.e_means = { epochs = 5 }under
[build.internal].The Rust trainer adapts the mini-batch updates from Faiss 1.16.0 and uses
VectorChord's RaBitQ for centroid assignment.
Paper · Faiss reference
Benchmark
On VectorChord 1.1.1 and PostgreSQL 17.11, five epochs of e-means
reduced
CREATE INDEXtime from 18.52 s to 13.43 s (27.5%).Centroid training took 4.46 s versus 9.04 s for Lloyd.
The workload used 167,689
halfvec(4096)embeddings, inner product,1,637 lists, 16 training threads, and 2,000 queries. Both configurations
were measured once on the same node. Recall@10 against exact float32
search differed by at most 0.23 percentage points across 64, 128, 256,
and 512 probes.
Validation
Separately verified on upstream
mainat0768657with PostgreSQL 17.11.make build,CREATE EXTENSION, five Rust tests, and three SQL regressionfiles passed. The new SQL cases cover index creation, nearest-neighbor
search, and rejection of
epochs = 0.