bhack
bhack
I've seen this but I don't understand how it is going to work in a multi-users setup with torchserve. Is it going to exchange the session/state forth - and back...
Interesting, but it could be harder to do it with a standard exported model which is internally stateful. So it seems we need to have something similar by design. In...
> @mreso could you please check if we can provide a SAM2 example for CX? It seems that users are going to have issues exporting the video mode for onnx:...
There are also other recent approaches like https://github.com/Vchitect/Latte
I don't understand why in https://github.com/alicevision/Meshroom/pull/2033 the calibration checkboard node is mandatory.
Can I ask you why you are still using the external flash attention? `torch.nn.functional.scaled_dot_product_attention` has already a flashattention-2 implementation: https://pytorch.org/docs/2.2/generated/torch.nn.functional.scaled_dot_product_attention.html
Probably we could monitor this https://github.com/pytorch/pytorch/pull/120642
> We also need evidence for the measure of inactivitiy, e.g. using devstats. Is there anyone finalizing this https://github.com/cncf/toc/issues/1139#issuecomment-1901093063
https://opensource.microsoft.com/blog/2024/02/06/on-device-training-training-a-model-in-browser/
Also fine-tuning other Meta foundational models would be nice like recent https://github.com/facebookresearch/segment-anything-2