Issues

• CUDA: reshape backward of a broadcast tensor only visits the input's storage, so the gradient of broadcast copies is lost

  chelsea0x3b/dfdx#933 · Filter by author nuxyel opened on Oct 20, 2026

• CPU: slice backward assigns into the input's gradient instead of accumulating (wrong gradients when a sliced tensor is also used elsewhere)

  chelsea0x3b/dfdx#932 · Filter by author nuxyel opened on Oct 20, 2026

• A pure Rust mini-language for sized vector types

  chelsea0x3b/dfdx#930 · Filter by author stefanmohl opened on Jul 27, 2025

• Basic operators moving tensor variables

  chelsea0x3b/dfdx#929 · Filter by author otavio-silva opened on Oct 20, 2024

• Safetensor panics instead of returning Err

  chelsea0x3b/dfdx#928 · Filter by author opfromthestart opened on Aug 24, 2024

• Tests failing on HEAD /w cargo 1.78.0

  chelsea0x3b/dfdx#926 · Filter by author twitchyliquid64 opened on Jul 17, 2024

• Kernels written in rust-gpu

  chelsea0x3b/dfdx#925 · Filter by author LegNeato opened on Apr 14, 2024
  1

• can not perform tensor operations (add/sub/mul/...) on tensors with non float datatype

  chelsea0x3b/dfdx#924 · Filter by author ghost opened on Apr 8, 2024

• Non-trainable parameters?

  chelsea0x3b/dfdx#921 · Filter by author hovinen opened on Mar 1, 2024
  2

• CUDA kernels missing __hmin and __hmax

  chelsea0x3b/dfdx#910 · Filter by author opfromthestart opened on Jan 15, 2024
  1

• OpenXLA Support

  chelsea0x3b/dfdx#908 · Filter by author vyeevani opened on Jan 12, 2024
  3

• Unable to build with old CUDA version (CUDA_COMPUTE_CAP = 52)

  chelsea0x3b/dfdx#907 · Filter by author nikvoid opened on Jan 9, 2024

• Add nn::AdaptiveAvgPool2D

  chelsea0x3b/dfdx#903 · Filter by author quietlychris opened on Dec 31, 2023

• Unclear how to handle error type in dfdx::nn::LoadFromNpz::load

  chelsea0x3b/dfdx#900 · Filter by author hovinen opened on Dec 11, 2023

• Add OUTPUT_PADDING to ConvTrans2D

  chelsea0x3b/dfdx#889 · Filter by author swfsql opened on Nov 15, 2023

• Send/Sync for Device

  chelsea0x3b/dfdx#888 · Filter by author blogle opened on Nov 10, 2023
  1

• Examples or resources for autodiff with 2 networks?

  chelsea0x3b/dfdx#880 · Filter by author Cobrand opened on Oct 30, 2023

• Device doesn't cover all traits from Cpu/Cuda

  chelsea0x3b/dfdx#866 · Filter by author jackerschott opened on Sep 8, 2023
  2

• Grouped Linear layers

  chelsea0x3b/dfdx#857 · Filter by author opfromthestart opened on Aug 25, 2023
  1

• TapeGlobal and thoughts on the tape variants

  chelsea0x3b/dfdx#843 · Filter by author emchristiansen opened on Aug 3, 2023
  3

• Bug: Incompatible uniqueness assumption causes crashes

  chelsea0x3b/dfdx#842 · Filter by author emchristiansen opened on Aug 3, 2023
  1

• The merge semantics for Arc<Mutex<OwnedTape<_, _>>> seem a bit unintuitive

  chelsea0x3b/dfdx#841 · Filter by author emchristiansen opened on Aug 2, 2023
  2

• Add Gradient scaler

  chelsea0x3b/dfdx#832 · Filter by author chelsea0x3b opened on Jul 26, 2023
  1

• Please add a split method, to iterate over subtensors of the given tensor

  chelsea0x3b/dfdx#825 · Filter by author emchristiansen opened on Jul 21, 2023
  7

• impl Default for Tensor?

  chelsea0x3b/dfdx#822 · Filter by author emchristiansen opened on Jul 20, 2023
  3