Search results

Split wikinightly.tasks to manage growing nightly runtime

Status: Open.

#594 In mikemccand/luceneutil;

· sgup432 opened on Jul 17, 2026

Ensure knnPerfTest.py uses new query vectors for warmup vs real test

Status: Open.

#591 In mikemccand/luceneutil;

· mikemccand opened on Jul 7, 2026

Add utility to measure vector search CPU cost sweeping across varying inputs like topK, vector count, dimension, segment count while preserving recall

Status: Open.

#590 In mikemccand/luceneutil;

· mikemccand opened on Jun 30, 2026

Upgrade to Cohere v4 Matryoshka Wikipedia embeddings?

Status: Open.

#584 In mikemccand/luceneutil;

· mikemccand opened on Jun 8, 2026

Add "partitipation ratio" measure (maybe just "condition") for vectors so we can see how helpful preconditioning would be

Status: Open.

#580 In mikemccand/luceneutil;

· mikemccand opened on May 11, 2026

Why is runNightlyKnn.py using inefficient non-optimized non-bulk Panama APIs?

Status: Open.

#579 In mikemccand/luceneutil;

· mikemccand opened on May 10, 2026

Implement NativeVectorizationProvider using BLAS library

Status: Open.

#573 In mikemccand/luceneutil;

· mikemccand opened on May 1, 2026

Should nightly benchmarks swap in a performant (native JMA) NativeVectorizationProvider?

Status: Open.

#572 In mikemccand/luceneutil;

· mikemccand opened on May 1, 2026

Understand hot CPUTime methods from nightly vectors index

Status: Open.

#571 In mikemccand/luceneutil;

· mikemccand opened on May 1, 2026

Record more JFR events like C2 compile / decompile, gc/heap, etc.

Status: Open.

#570 In mikemccand/luceneutil;

· mikemccand opened on May 1, 2026

Expose SIMD CPU counters over time (-I interval option to perf)

Status: Open.

#569 In mikemccand/luceneutil;

· mikemccand opened on Apr 24, 2026

Enable measuring recall loss due solely to quantization

Status: Open.

#567 In mikemccand/luceneutil;

· mikemccand opened on Apr 17, 2026