GPU usage

[RobertoLando]

I'm using a few generated scripts on an EC2 instance of type g4dn.xlarge, with Tesla T4 GPU

When running the workflow from the ComfyUI interface, I see traced in logs:
Device: cuda:0 Tesla T4 : cudaMallocAsync
while if I run the generated python script I see:
Device: cuda:0 Tesla T4 : native

The script produces similar result, but it's a lot slower. I tryied to add --cuda-malloc to the command line, but nothing changes. Any clue?

btw: great extension!

[RobertoLando]

Found a solution: in the python script generated by this extension you need to add import cuda_malloc at the beginning (with some code to verify if it is supported) and move the import torch statement from the beginning to the first line after def main():

It would be useful to generate code already in this form

[min-star]

Found a solution: in the python script generated by this extension you need to add import cuda_malloc at the beginning (with some code to verify if it is supported) and move the import torch statement from the beginning to the first line after def main():

It would be useful to generate code already in this form

i add the code,but the cost time is same with not add