Spaces:
Running
on
L40S
gradio tqdm abnormal display
When space is launched for the first time, it will compile the model once. This compilation takes less than an hour, and the displayed screen is your "error screen." This is normal. However, from the second time onwards, as long as space is not shut down, it will continue to infer at a faster speed. You can disable this optimization by removing this code
pipe.transformer = torch.compile(pipe.transformer, mode="max-autotune", fullgraph=True)
However, doing so requires an additional 30 seconds for each inference compared to after compilation
When space is launched for the first time, it will compile the model once. This compilation takes less than an hour, and the displayed screen is your "error screen." This is normal. However, from the second time onwards, as long as space is not shut down, it will continue to infer at a faster speed. You can disable this optimization by removing this code
pipe.transformer = torch.compile(pipe.transformer, mode="max-autotune", fullgraph=True)
However, doing so requires an additional 30 seconds for each inference compared to after compilation
You're so great, thank you very much!