You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Test FramePack-F1 on different versions of Torch
#798
Measured on a triple run of a 5-second video with the prompt "The four girls dance gracefully, with clear movements, full of charm." :) Each video's seed changed to +1
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Environment:
Python 3.12.10(venv)/CUDA 12.8
GF 4080 16Gb (+32Gb shared=48Gb), driver 581.57 | Ryzen 5 5600X/RAM 64Gb
teacache on
Measured on a triple run of a 5-second video with the prompt "The four girls dance gracefully, with clear movements, full of charm." :) Each video's seed changed to +1
Torch 2.6+CUDA 12.6
torch==2.6.0+cu126
torchvision==0.21.0+cu126
xformers==0.0.29.post3
sageattention==2.2.0+cu126torch2.6.0.post3 (fork, https://github.com/woct0rdho/SageAttention/releases)
VRAM usage: 13.8Gb
Speed: 2.87-2.92s/it
Torch 2.7+CUDA 12.8
pip --python .\venv install --upgrade torch==2.7.0 torchvision==0.22.0 torchaudio==2.7.0 numpy==1.26.4 xformers==0.0.30 sageattention --index-url https://download.pytorch.org/whl/cu128 --extra-index-url https://pypi.org/simpletorch==2.7.0+cu128
torchvision==0.22.0+cu128
xformers==0.0.30
sageattention==1.0.6 (latest offical)
VRAM usage: 13.4Gb
Speed: 2.92-3.07s/it
Torch 2.8.0 + CUDA 12.8
torch==2.8.0+cu128
torchvision==0.23.0+cu128
xformers==0.0.32.post2
sageattention==2.2.0+cu128torch2.8.0.post3 (fork, https://github.com/woct0rdho/SageAttention/releases)
VRAM usage: 14Gb
Speed: 2.70-2.82 s/it
Torch 2.9.1+CUDA 12.8
pip --python .\venv install --upgrade torch torchvision torchaudio numpy==1.26.4 xformers sageattention --index-url https://download.pytorch.org/whl/cu128 --extra-index-url https://pypi.org/simpletorch==2.9.1+cu128
torchvision==0.24.1+cu128
xformers==0.0.33.post2
sageattention==1.0.6 (latest offical)
VRAM usage: 13.9Gb on first pass (25 iteration), 27Gb on subsequent passes
Speed: 3.14s/it on first pass, then interrupted
It's likely due to the version of Torch that Out-Of-Memory occurs in ComfyUI. VRAM is not freed after each pass.
*numpy==1.26.4 required for latest releases open-cv (image processing) and it is advisable to add a line to requirements.txt
*Add Torch 2.8.0 + CUDA 12.8
All reactions