Repository navigation

perf(pt_expt): fuse the mixing-stack per-layer backward chain (level-2 tuned table) - #6060

Open

long-yi-2019 wants to merge 11 commits into

deepmodeling:masterfrom

long-yi-2019:pr/stack-bwd-fused

Open

perf(pt_expt): fuse the mixing-stack per-layer backward chain (level-2 tuned table)#6060
long-yi-2019 wants to merge 11 commits into
deepmodeling:masterfrom
long-yi-2019:pr/stack-bwd-fused

Commits on Oct 8, 2026
- perf(pt): fuse the mixing-stack per-layer backward chain (DP_STACK_BWD_FUSED_INFER)
  longjunhao
  committed
  a3dcd97
- [pre-commit.ci] auto fixes from pre-commit.com hooks
  pre-commit-ci[bot]
  committed
  4455e67
Commits on Oct 16, 2026
- test(pt): exercise the stack-backward fallback guards with the flag enabled
  longjunhao
  committed
  ad9af59
Commits on Oct 17, 2026
- Merge branch 'master' into pr/stack-bwd-fused
  OutisLi
  authored
  633e87c
Commits on Oct 18, 2026
- feat(pt): select the fused mixing-stack backward by tuned tile entry
  longjunhao
  committed
  fa5e98e
- test(pt): retarget the fused stack-backward tests at the tuned selection
  longjunhao
  committed
  8476318
- fix(pt): time fused sweep candidates under no_grad and assert the selection in tests
  longjunhao
  committed
  2e8e6f3
Commits on Oct 21, 2026
- fix(pt_expt): make stack_bwd_fused tuning trustworthy and dual-backend
  longjunhao
  committed
  5df6813
- [pre-commit.ci] auto fixes from pre-commit.com hooks
  pre-commit-ci[bot]
  committed
  ef2eaf7
Commits on Oct 22, 2026
- fix(pt): keep the freeze tuning seam import-safe and backend-neutral
  longjunhao
  committed
  f524062
Commits on Oct 25, 2026
- fix(pt_expt): give the fused stack backward a non-aliasing output buffer
  longjunhao
  committed
  8a61fbb