li-lizhe 7d6a98bc62 fix(framepack): use text_mask.device instead of hardcoded cuda ...

get_cu_seqlens creates a cu_seqlens tensor with device="cuda",
which crashes on non-CUDA devices (e.g. Ascend NPU raises
"Torch not compiled with CUDA enabled"). Create it on the same
device as the input text_mask instead, which is device-agnostic.

2026-09-12 09:08:48 +08:00

..

bucket_tools.py

add builtin framepack

2025-07-08 15:47:07 -04:00

clip_vision.py

add builtin framepack

2025-07-08 15:47:07 -04:00

dit_common.py

add builtin framepack

2025-07-08 15:47:07 -04:00

hunyuan_video_packed.py

fix(framepack): use text_mask.device instead of hardcoded cuda

2026-09-12 09:08:48 +08:00

hunyuan.py

Do not ignore offload with FramePack

2025-07-13 20:41:55 +03:00

k_diffusion_hunyuan.py

add builtin framepack

2025-07-08 15:47:07 -04:00

thread_utils.py

add builtin framepack

2025-07-08 15:47:07 -04:00

uni_pc_fm.py

Revert "much stricter ruff linting"

2026-05-11 08:13:57 +02:00

utils.py

modernize typing

2026-02-19 09:15:37 +01:00

wrapper.py

add builtin framepack

2025-07-08 15:47:07 -04:00