Use `flashdreams.accelerated` in Wan family models - #487

Closed

fangjunzhou-nv wants to merge 12 commits into

NVIDIA:mainfrom

fangjunzhou-nv:dev/fangjun/flashdreams-accelerated-wan-integrations

Commits on Aug 18, 2026

Environment config
fangjunzhou-nv
committed
74dba42
Add pytest-benchmark and nvtx to project dependency
fangjunzhou-nv
committed
5f81148
Add nvtx annotation to pipeline execution
fangjunzhou-nv
committed
b35697a
Add GB300 support to omnidreams CUDA DiT backend
fangjunzhou-nv
committed
df94e1f
Add flashdreams.accelerated v0 API
fangjunzhou-nv
committed
6235c8b
Add benchmark for flashdreams.accelerated
fangjunzhou-nv
committed
558e309
Add Triton MHA backend for Wan recipe
fangjunzhou-nv
committed
779d01e
Add Triton backend Wan DiT for lingbot and Wan 2.1
fangjunzhou-nv
committed
7b8f4ff
Add Triton backend MHA for omnidreams
fangjunzhou-nv
committed
d0c0105
Add benchmark scripts
fangjunzhou-nv
committed
3077d4d
Ignore neovim config
fangjunzhou-nv
committed
e029437

Commits on Aug 19, 2026

Removed Torch MHA and simplified Triton test
fangjunzhou-nv
committed
2de93ae