cpuhrsch

Results 178 comments of cpuhrsch

Functorch Usecase Jacobian computation

We now have support for [NestedTensors](https://pytorch.org/docs/master/nested.html) in core. That means you could construct a NestedTensor via ``` torch.nested_tensor([ torch.rand(batch_size, in_channels, feature_shape, feature_shape), torch.rand(batch_size, in_channels, feature_shape-10, feature_shape-10), ]) ``` Having said...

Functorch Usecase Jacobian computation

cc @jbschlosser

Functorch Usecase Jacobian computation

@zou3519 - that would be 5 dimensional. conv2d requires 4-dim inputs, so it wouldn't be able to accept it. I think in general vmap support for NestedTensor is a great...

csr <-> csc, csc <-> csc, bsr <-> bsc, bsc <-> bsc, bsr <-> bsr conversions

@pytorchbot merge

csr <-> csc, csc <-> csc, bsr <-> bsc, bsc <-> bsc, bsr <-> bsr conversions

@pytorchbot merge -f "failure is unrelated"

Enable dense x bsc mm/addmm

@pytorchbot merge

Enable dense x bsc mm/addmm

@pytorchbot merge

resize_as_sparse support all compressed layouts

@pytorchbot merge

Get flash_attn to compile for CUDA 11.6 linux nightly build

@drisspg - as long as the compiler is modern enough it should be possible. We'll see if we need to guard against certain compiler versions.

Get flash_attn to compile for CUDA 11.6 linux nightly build

I'm worried about the CUDA context size due to this change. cc @ngimel