Comparing
claude/optimize-fsst-compression-KgJdu
(280a766
) withdevelop
(b921999
)-84%
- Improvement
- 14
- Regression
- 12
- Untouched
- 933
- New
- 44
- Archived
- 1565
Benchmarks
2568 total
train_compressor[(10000, 16, 8)]
encodings/fsst/benches/fsst_compress.rs-84%
4.8 ms30.8 ms
train_compressor[(10000, 16, 4)]
encodings/fsst/benches/fsst_compress.rs-84%
4.8 ms30.1 ms
train_compressor[(10000, 64, 8)]
encodings/fsst/benches/fsst_compress.rs-83%
4.7 ms27.1 ms
train_compressor[(10000, 64, 4)]
encodings/fsst/benches/fsst_compress.rs-81%
4.7 ms25 ms
train_compressor[(10000, 4, 4)]
encodings/fsst/benches/fsst_compress.rs-76%
4.9 ms20.7 ms
train_compressor[(10000, 4, 8)]
encodings/fsst/benches/fsst_compress.rs-76%
5.2 ms21.9 ms
train_compressor[(1000, 64, 4)]
encodings/fsst/benches/fsst_compress.rs-75%
4.2 ms17.1 ms
train_compressor[(1000, 64, 8)]
encodings/fsst/benches/fsst_compress.rs-74%
4.3 ms16.7 ms
train_compressor[(1000, 16, 4)]
encodings/fsst/benches/fsst_compress.rs-56%
4.2 ms9.6 ms
train_compressor[(1000, 16, 8)]
encodings/fsst/benches/fsst_compress.rs-47%
4.2 ms8 ms
train_compressor[(1000, 4, 8)]
encodings/fsst/benches/fsst_compress.rs-44%
2 ms3.6 ms
train_compressor[(1000, 4, 4)]
encodings/fsst/benches/fsst_compress.rs-42%
1.9 ms3.2 ms
chunked_canonicalize_into[(1000, 50, 8, 64)]
encodings/fsst/benches/fsst_compress.rs+43%
10.5 ms7.3 ms
chunked_into_canonical[(1000, 50, 8, 64)]
encodings/fsst/benches/fsst_compress.rs+42%
10.8 ms7.6 ms
chunked_canonicalize_into[(1000, 100, 16, 16)]
encodings/fsst/benches/fsst_compress.rs+38%
20.5 ms14.8 ms
chunked_into_canonical[(1000, 100, 16, 16)]
encodings/fsst/benches/fsst_compress.rs+37%
20.8 ms15.2 ms
chunked_canonicalize_into[(1000, 50, 8, 16)]
encodings/fsst/benches/fsst_compress.rs+33%
8.9 ms6.7 ms
chunked_into_canonical[(1000, 50, 8, 16)]
encodings/fsst/benches/fsst_compress.rs+32%
9.1 ms6.9 ms
chunked_canonicalize_into[(1000, 100, 16, 64)]
encodings/fsst/benches/fsst_compress.rs+25%
23.9 ms19.1 ms
chunked_into_canonical[(1000, 100, 16, 64)]
encodings/fsst/benches/fsst_compress.rs+25%
24.2 ms19.3 ms
eq_canonicalize_low_match
encodings/fsst/benches/fsst_url_compare.rs+17%
30.9 ms26.4 ms
eq_canonicalize_high_match
encodings/fsst/benches/fsst_url_compare.rs+17%
31.3 ms26.8 ms
chunked_canonicalize_into[(1000, 50, 8, 4)]
encodings/fsst/benches/fsst_compress.rs+14%
7.8 ms6.8 ms
like_substr_low_match
encodings/fsst/benches/fsst_url_compare.rs+13%
38.2 ms33.7 ms
chunked_into_canonical[(1000, 50, 8, 4)]
encodings/fsst/benches/fsst_compress.rs+13%
8 ms7.1 ms
Commits
Click on a commit to change the comparison rangeBase
develop
b921999
-85.68%
feat(fsst): improve training convergence with stratified sampling and more generations
9a2c09d
3 months ago
by claude+0.01%
perf(fsst): hoist PHT lookup for instruction-level parallelism in compress_word
ec7894e
3 months ago
by claude+1.4%
feat(fsst): remove min_count pre-filter in favor of gain-based selection
280a766
3 months ago
by claude