Is there any plan to support fastercache?

[songh11]

Expect to support fastercache, https://github.com/Vchitect/FasterCache

[a-r-r-o-w]

Thanks for the suggestion 🤗 Yes, we will be looking into the new caching methods starting next week - PAB, AdaCache, FasterCache, TeaCache, etc! Just ideating and finalizing the design on how to best integrate these

[songh11]

Thanks for the suggestion 🤗 Yes, we will be looking into the new caching methods starting next week - PAB, AdaCache, FasterCache, TeaCache, etc! Just ideating and finalizing the design on how to best integrate these

Thank you for the response❤️. Additionally these methods involve dynamic models, may I ask if these are compatible with torch.compile?

[a-r-r-o-w]

Initially, full compatibility with max-autotune will probably not be possible, because of dynamic execution paths due to data dependant branching, like you mentioned. But wherever possible, we can try to stay within the torch.compile guidelines

[songh11]

Initially, full compatibility with max-autotune will probably not be possible, because of dynamic execution paths due to data dependant branching, like you mentioned. But wherever possible, we can try to stay within the torch.compile guidelines

Thank you, looking forward to the future development of cache.