Ybzhang2001
Ybzhang2001
Guess Playground doesn't inherit the weights of the SD-XL pre-trained model, despite using the same architecture.
Thanks for the response! Please notify me when the corrected version is released, thank you.
For comparison, the hunyuanvideo-13B finetune script in V0 (e.g., finetune_hunyuan.sh) seems to support input shapes like 1280×720×125 by default. Moreover, I have successfully trained with the shape 512×512×121 and bf16...
> it should work, could you share the scripts you are using? Did you turn on SP? NUM_GPUS=8 torchrun --nnodes 1 --nproc_per_node $NUM_GPUS \ fastvideo/v1/training/wan_training_pipeline.py\ --model_path /root/highspeedstorage/Data01/weights/Wan2.1-T2V-14B-Diffusers \ --inference_mode False\...
> Could you try add `--enable_gradient_checkpointing_type "full" ` (May need to pull first to ensure it support gradient_checkpointing您可以尝试添加 `--enable_gradient_checkpointing_type "full" ` 吗（可能需要先拉取以确保它支持梯度检查点 Thanks for your reply! But `--enable_gradient_checkpointing_type "full"` has...