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/.

This commit is contained in:

Linz

2026-07-20 16:50:59 +00:00

parent 26b071c6e1

commit 00902e6c67

12 changed files with 502 additions and 6 deletions

.gitignore

+2 -1

View File

@@ -1,2 +1,3 @@
 runs/
 runs/*
 !runs/shape/
 __pycache__/