masahi commented on issue #10223:
URL: https://github.com/apache/tvm/issues/10223#issuecomment-1052837280
https://github.com/apache/tvm/tree/main/gallery/how_to/tune_with_autoscheduler
has e2e examples of using the auto scheduler. But yeah, I don't expect it to
beat cuDNN, unless cuDNN
masahi commented on issue #10223:
URL: https://github.com/apache/tvm/issues/10223#issuecomment-1051988687
oops good find! Can you send a PR? I can quickly merge it (if I do it you
need to wait until next week).
> Sadly I'm still just a few FPS shy of my performance target so I'll
masahi commented on issue #10223:
URL: https://github.com/apache/tvm/issues/10223#issuecomment-1039219766
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe,
masahi commented on issue #10223:
URL: https://github.com/apache/tvm/issues/10223#issuecomment-1039788394
Are you sure you can run your PT model on cuda via cudnn? I'm getting
`CUDNN_STATUS_BAD_PARAM` when trying to run your model with cudnn. A WIP branch
masahi commented on issue #10223:
URL: https://github.com/apache/tvm/issues/10223#issuecomment-1039219766
Yeah, cuda backend doesnt support groups. You can try the cpu backend to
verify the result. If you are ok with using cudnn, I can quickly enable groups
support for cudnn conv
masahi commented on issue #10223:
URL: https://github.com/apache/tvm/issues/10223#issuecomment-1037881211
Fixed in https://github.com/apache/tvm/pull/10235. You shouldn't be using
`torch.jit.optimize_for_inference`, it does no good for us and it even
introduces `aten::conv2d` etc that we
masahi commented on issue #10223:
URL: https://github.com/apache/tvm/issues/10223#issuecomment-1037881211
Fixed in https://github.com/apache/tvm/pull/10235. You shouldn't be using
`torch.jit.optimize_for_inference`, it does no good for us and it even
introduces `aten::conv2d` etc that we
masahi commented on issue #10223:
URL: https://github.com/apache/tvm/issues/10223#issuecomment-1036489478
Thanks, yes `conv2d_transpose` with group was only recently fixed and
supported in https://github.com/apache/tvm/pull/9465. I think we haven't
updated our PyTorch frontend to benefit
