3 Commits

Author	SHA1	Message	Date
zhoulin	c8a42b66ee	Move trace shape sets to top-level shape/ and frame ab_fold_test.sh as A/B reference example	2026-07-20 16:53:14 +00:00
zhoulin	00902e6c67	Default USE_FLAGTUNE to 0 and add fold_scale A/B script with trace shapes ... Plain autotune is the common case (fast, same-config A/B); FlagTune's exhaustive first-run search is now opt-in via USE_FLAGTUNE=1. ab_fold_test.sh automates the FLAGGEMS_MXFP4_FOLDSCALE=0/1 record-replay comparison over the four DeepSeek-V4-Flash trace shape sets, which are now tracked under runs/shape/.	2026-07-20 16:50:59 +00:00
zhoulin	8ace0d790c	Add FlagGems single-op perf benchmark harness ... pytest plugin set + driver script for compiler A/B perf comparison: - reproducible runs: fixed seed, yaml-driven shapes, autotune record/replay - per-shape ttgir dump of actually-used variants with readable naming - cudagraph-based timing with documented fallback semantics Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>	2026-07-16 03:52:07 +00:00