13b distilled fp8 crashes with no error message

[rickyars]

i have a 4080 super. everything seems to be installed correctly. however, every time i try to run the example workflow comfy crashes during the first round of ltxv sampling. any suggestions?

Skip steps sigma threshold:  0.9970000000000002
Cfg star rescale:  True
Sigma list:  [1.0, 0.9933, 0.985, 0.9767, 0.9008, 0.618]
Cfg list:  [1.0, 1.0, 1.0, 1.0, 1.0, 1.0]
Stg scale list:  [0.0, 0.0, 0.0, 0.0, 0.0, 0.0]
Stg rescale list:  [1.0, 1.0, 1.0, 1.0, 1.0, 1.0]
Stg layers indices list:  [[25], [35], [35], [42], [42], [42]]
optional_cond_images shape torch.Size([1, 512, 768, 3])
Requested to load VideoVAE
loaded partially 322.49140624999995 322.37719917297363 0
Requested to load LTXV
loaded partially 435.94062499999995 435.8594970703125 0
  0%|                                                                                            | 0/8 [00:00<?, ?it/s]

at this point - i'm back to the terminal and have to restart comfy

[rickyars]

i've updated comfyui i've install q8_kernel

[CaptainArmenia]

same here, using a desktop 4090

[rickyars]

using Claude, i was able to diagnose the promblem. even though q8_kernel claimed to install correctly, the import was failing. we only figured this out when trying to use the q8_kernel directly in python.

so, i gave up on the fp8 version and the distilled seems to be running just fine.

NOTE: i'm on windows

[rtaskf]

This happens for me with both the distilled and non-distilled model. Did everything as instructed. Out of ideas.

[michaellightricks]

Is there any error message when it crashes?

[rickyars]

there were not error messages because the cause of the failure is on import of the q8_kernel. i ended up giving up.