Spaces:
Running
ZeroGPU Spaces go brrr with ahead-of-time compilation
Community discussion following up on this blog post, where we showed how PyTorch’s ahead-of-time compilation techniques can be used to speed up ZeroGPU applications, with examples applied for Flux dev, Flux Kontext, Wan 2.2 and Qwen Image Edit.
These techniques are not limited to just these models! We encourage you to give these techniques a try and provide us with feedback
This is magic , i am playing around but what i don't understand is https://huggingface.co/spaces/rahul7star/wan2-2-T2V-EXP/blob/main/optimization.py#L43. where is this 1500 ( 25 mins space time ) gets executed . I was avoiding to ask this cause who knows if you guys block this and my magic moments will be gone :). if you throw that 1500 to user then all space time will be over in cold start only
Checking your article .. if this is the way to go then perhaps it has be a part of library :) Great work