cmoyates
|
d13ba1639d
|
feat: add optimization toggle flags + benchmark matrix script
use_sdpa, stage_gc flags propagated through backbone/model/pipeline.
bench_optimizations.py runs exhaustive/ablation/key sweep of all 6
toggles (slim, stage_gc, sdpa, bf16, fused_decode, gpu_preprocess)
with latency + peak memory reporting.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
|
2026-03-09 17:11:28 -02:30 |
|