Skip to content

Experiment: price compression for FineWeb predicates#8863

Closed
connortsui20 wants to merge 1 commit into
ct/compressor-cost-modelfrom
codex/predicate-heavy-cost
Closed

Experiment: price compression for FineWeb predicates#8863
connortsui20 wants to merge 1 commit into
ct/compressor-cost-modelfrom
codex/predicate-heavy-cost

Add query-aware compression cost experiment

78df2bf
Select commit
Loading
Failed to load commit list.
CodSpeed HQ / CodSpeed Performance Analysis succeeded Jul 20, 2026 in 0s

Performance Gate Passed

⚠️ Unknown Walltime execution environment detected

Using the Walltime instrument on standard Hosted Runners will lead to inconsistent data.

For the most accurate results, we recommend using CodSpeed Macro Runners: bare-metal machines fine-tuned for performance measurement consistency.

⚡ 6 improved benchmarks
✅ 1692 untouched benchmarks
🆕 12 new benchmarks

Performance Changes

Mode Benchmark BASE HEAD Efficiency
WallTime cuda/bitpacked_u8/unpack/3bw[100M] 646.3 µs 300.8 µs ×2.1
WallTime cuda/bitpacked_patched_u8/unpack/1%[100M] 768.9 µs 382.3 µs ×2
WallTime cuda/bitpacked_patched_u8/unpack/10%[100M] 828.6 µs 477.4 µs +73.57%
WallTime cuda/bitpacked_patched_u16/unpack/1%[100M] 712.3 µs 596.1 µs +19.5%
WallTime cuda/bitpacked_patched_u16/unpack/10%[100M] 831 µs 715.5 µs +16.15%
WallTime cuda/bitpacked_u16/unpack/5bw[100M] 634.8 µs 557 µs +13.95%
🆕 Simulation canonical_decode N/A 610.6 ns N/A
🆕 Simulation canonical_eq N/A 13.6 ms N/A
🆕 Simulation canonical_like N/A 23.1 ms N/A
🆕 Simulation dict_decode N/A 42.1 ms N/A
🆕 Simulation dict_eq N/A 1.2 ms N/A
🆕 Simulation dict_like N/A 65.1 ms N/A
🆕 Simulation fsst_decode N/A 25.3 ms N/A
🆕 Simulation fsst_eq N/A 1.9 ms N/A
🆕 Simulation fsst_like N/A 12.1 ms N/A
🆕 Simulation like_winner_decode N/A 25.3 ms N/A
🆕 Simulation like_winner_eq N/A 1.9 ms N/A
🆕 Simulation like_winner_like N/A 12.1 ms N/A

Tip

Curious why this is faster? Comment @codspeedbot explain why this is faster on this PR, or directly use the CodSpeed MCP with your agent.


Comparing codex/predicate-heavy-cost (78df2bf) with ct/compressor-cost-model (43f6c42)

Open in CodSpeed