dfdx Files

What's Changed

• Fix bug in gather cuda kernel by @nkoppel in https://github.com/coreylowman/dfdx/pull/588
• feat(device): introduce AutoDevice type by @kakoc in https://github.com/coreylowman/dfdx/pull/579
• Use Recursive Macros to Implement Shape Operation Traits. by @nkoppel in https://github.com/coreylowman/dfdx/pull/583
• Add ToDtype tensor operation by @nkoppel in https://github.com/coreylowman/dfdx/pull/582
• Using 128 threads by default for cuda kernels by @coreylowman in https://github.com/coreylowman/dfdx/pull/599
• Add Slice tensor operation. by @nkoppel in https://github.com/coreylowman/dfdx/pull/602
• Optimizing conv kernels a bit by @coreylowman in https://github.com/coreylowman/dfdx/pull/605
• feat: add upper/lower triangles (tril and triu) allocations by @Alexandcoats in https://github.com/coreylowman/dfdx/pull/568
• Adds Tensor::roll by @coreylowman in https://github.com/coreylowman/dfdx/pull/608
• Using multiple streams for matmul with cuda by @coreylowman in https://github.com/coreylowman/dfdx/pull/610
• Fix no-std support by @Alexandcoats in https://github.com/coreylowman/dfdx/pull/615
• Adds matrixmultiply/std to std feature by @kstavro in https://github.com/coreylowman/dfdx/pull/618
• Implement concat for usize arrays; add concat to Device. by @nkoppel in https://github.com/coreylowman/dfdx/pull/621
• Allow conv2d and pool2d to use dynamic dimensions for width and height. by @nkoppel in https://github.com/coreylowman/dfdx/pull/620
• Switch to using nvcc --list-gpu-code for build.rs compute_cap by @quietlychris in https://github.com/coreylowman/dfdx/pull/619
• Fix bug in reshape on cuda by @nkoppel in https://github.com/coreylowman/dfdx/pull/622
• Don't always do try_min in pool_global.rs by @nkoppel in https://github.com/coreylowman/dfdx/pull/623
• Revert "Switch to using nvcc --list-gpu-code for build.rs compute_cap… by @coreylowman in https://github.com/coreylowman/dfdx/pull/624
• Adds restrided in favor of get_unstrided_index -> get_strided_index by @coreylowman in https://github.com/coreylowman/dfdx/pull/628
• Combines multiple calls to get_strided_index into a single loop by @coreylowman in https://github.com/coreylowman/dfdx/pull/629
• Reducing number of buffers sent to cuda for some operations by @coreylowman in https://github.com/coreylowman/dfdx/pull/611
• Optimizing conv2d more by @coreylowman in https://github.com/coreylowman/dfdx/pull/631
• Add ability to include smaller last batch by @nkoppel in https://github.com/coreylowman/dfdx/pull/632
• Upscale2D and ConvTrans2d by @opfromthestart in https://github.com/coreylowman/dfdx/pull/603
• impl Dtype for all Unit types except bool by @coreylowman in https://github.com/coreylowman/dfdx/pull/635
• Allow convtrans2d to use dynamic dimensions by @nkoppel in https://github.com/coreylowman/dfdx/pull/639
• JIT compiling kernel for to_dtype & reshape by @coreylowman in https://github.com/coreylowman/dfdx/pull/634
• Optimize conv transpose kernels to do same thing as conv by @coreylowman in https://github.com/coreylowman/dfdx/pull/641
• Reworking crate level documentation by @coreylowman in https://github.com/coreylowman/dfdx/pull/644
• Adds synchronize to DeviceStorage by @coreylowman in https://github.com/coreylowman/dfdx/pull/645
• adding usize dtype to cuda_kernel by @zojeda in https://github.com/coreylowman/dfdx/pull/648
• Add PReLU and LeakyReLU by @opfromthestart in https://github.com/coreylowman/dfdx/pull/586
• Moving logsumexp normalization off of graph by @coreylowman in https://github.com/coreylowman/dfdx/pull/652
• Adding CmpKernels to Device, more documentation by @coreylowman in https://github.com/coreylowman/dfdx/pull/653
• Removing bounds checking from cpu conv kernel folding by @coreylowman in https://github.com/coreylowman/dfdx/pull/650
• Allow upscale2d to use dynamic dimensions by @nkoppel in https://github.com/coreylowman/dfdx/pull/654
• Adding integration test for resnet18 by @coreylowman in https://github.com/coreylowman/dfdx/pull/655
• Removing some un-necessary blanket impls by @coreylowman in https://github.com/coreylowman/dfdx/pull/656
• Fixes conv transpose stride bug, adds more docs to upscale2d by @coreylowman in https://github.com/coreylowman/dfdx/pull/658
• Some QOL fixes by @opfromthestart in https://github.com/coreylowman/dfdx/pull/659
• Optimizing softmax & log_softmax by @coreylowman in https://github.com/coreylowman/dfdx/pull/660
• Reuse f(x) for unary operations when possible. by @coreylowman in https://github.com/coreylowman/dfdx/pull/661
• Allocating gradients in backward op by @coreylowman in https://github.com/coreylowman/dfdx/pull/663
• Adds Tensor::recip (1 / x) by @coreylowman in https://github.com/coreylowman/dfdx/pull/665
• Reshape layer by @opfromthestart in https://github.com/coreylowman/dfdx/pull/666
• Re-using tensor storage when possible by @coreylowman in https://github.com/coreylowman/dfdx/pull/664
• Adds cudnn feature flag. Removes "test-cuda" feature flag. Using cuDNN for convolutions by @coreylowman in https://github.com/coreylowman/dfdx/pull/651
• Always attempting allocation reuse during inference by @coreylowman in https://github.com/coreylowman/dfdx/pull/673
• Clarify reshape behavior in docs by @coreylowman in https://github.com/coreylowman/dfdx/pull/674
• Have SplitInto keep tapes of each head seperate by @nkoppel in https://github.com/coreylowman/dfdx/pull/676
• Using arch option in nvrtc by @coreylowman in https://github.com/coreylowman/dfdx/pull/675

New Contributors

• @kakoc made their first contribution in https://github.com/coreylowman/dfdx/pull/579
• @quietlychris made their first contribution in https://github.com/coreylowman/dfdx/pull/619
• @opfromthestart made their first contribution in https://github.com/coreylowman/dfdx/pull/603
• @zojeda made their first contribution in https://github.com/coreylowman/dfdx/pull/648

Full Changelog: https://github.com/coreylowman/dfdx/compare/v0.11.0...v0.11.1