[Bug]

[yomi1117]

Environment

The Environment is right.

Describe the bug

OOM...I know video distillation needs many GPUS. I used 8 A800 to run the distillation training of Hunyuan, but I got OOM error. How many GPUs we need to run the distill...

Reproduction

None

[foreverpiano]

8 GPUs are enough. Can you share with your script?

[jzhang38]

I just updated the script for 8 GPUs and can you try it here:

FastVideo/scripts/distill/distill_hunyuan.sh

Line 50 in dd75ee8

# If you do not have 32 GPUs and to fit in memory, you can: 1. increase sp_size. 2. reduce num_latent_t

[yomi1117]

Thank u for your help. I'm curious about how many GPUS and the config to train FastHunyuan~ Thank u !

[jzhang38]

This should be the exact command to reproduce FastHunyuan:

FastVideo/scripts/distill/distill_hunyuan.sh

Line 7 in dd75ee8

torchrun --nnodes 4 --nproc_per_node 8\