Rattus 8922c21c9e flux: Do the xq and xk ropes one at a time ...

This was doing independendent interleaved tensor math on the q and k
tensors, leading to the holding of more than the minimum intermediates
in VRAM. On a bad day, it would VRAM OOM on xk intermediates.

Do everything q and then everything k, so torch can garbage collect
all of qs intermediates before k allocates its intermediates.

This reduces peak VRAM usage for some WAN2.2 inferences (at least).

2025-09-16 22:53:31 +10:00

..

controlnet.py

Make flux controlnet work with sd3 text enc. (#8599)

2025-06-19 18:50:05 -04:00

layers.py

Enable Runtime Selection of Attention Functions (#9639)

2025-09-12 18:07:38 -04:00

math.py

flux: Do the xq and xk ropes one at a time

2025-09-16 22:53:31 +10:00

model.py

Enable Runtime Selection of Attention Functions (#9639)

2025-09-12 18:07:38 -04:00

redux.py

Support new flux model variants.

2024-11-21 08:38:23 -05:00