coreylowman/dfdx

Issues

• Basic operators moving tensor variables

  #929 opened 2 months ago by otavio-silva
  0

• Safetensor panics instead of returning Err

  #928 opened 4 months ago by opfromthestart
  0

• Kernels written in rust-gpu

  #925 opened 8 months ago by LegNeato
  1

• Non-trainable parameters?

  #921 opened 10 months ago by hovinen
  2

• Tests failing on HEAD /w cargo 1.78.0

  #926 opened 5 months ago by twitchyliquid64
  0

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

  #924 opened 8 months ago by lucomotive
  0

• Consider helpers for accessing tensors from tuples and input wrappers

  #878 opened 10 months ago by swfsql
  1

• Run tests with miri

  #897 opened 10 months ago by swfsql
  1

• Reduce test sizes

  #898 opened 10 months ago by swfsql
  1

• Add `Prodigy` optimizer

  #894 opened 10 months ago by swfsql
  1

• How does one update one model from another model?

  #905 opened 10 months ago by hovinen
  1

• OpenXLA Support

  #908 opened a year ago by vyeevani
  3

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

  #825 opened a year ago by emchristiansen
  7

• CUDA kernels missing __hmin and __hmax

  #910 opened a year ago by opfromthestart
  1

• `TapeGlobal` and thoughts on the tape variants

  #843 opened a year ago by emchristiansen
  3

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

  #907 opened a year ago by nikvoid
  0

• Add `nn::AdaptiveAvgPool2D`

  #903 opened a year ago by quietlychris
  0

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

  #900 opened a year ago by hovinen
  0

• Split `TryConcatAlong` into different traits

  #891 opened a year ago by swfsql
  0

• Replace explicit features and paths on generated code

  #885 opened a year ago by swfsql
  0

• Add `OUTPUT_PADDING` to `ConvTrans2D`

  #889 opened a year ago by swfsql
  0

• Send/Sync for Device

  #888 opened a year ago by blogle
  1

• Question / clarification regarding heap allocations

  #879 opened a year ago by emchristiansen
  2

• Bug: `Sequential` macro provide `forward_mut` as `forward`

  #881 opened a year ago by uselessgoddess
  0

• Examples or resources for autodiff with 2 networks?

  #880 opened a year ago by Cobrand
  0

• Deriving Shape for C like enums

  #856 opened a year ago by ZacharyKamerling
  1

• Is there an equivalent to torch.masked_select?

  #865 opened a year ago by jackerschott
  1

• Unnecessary loss of precision when computing loss functions

  #872 opened a year ago by ariasanovsky
  2

• trait TryConcatAlong not satisfied when using constants

  #874 opened a year ago by opfromthestart
  1

• Different results when CPU feature is on vs off

  #871 opened a year ago by jafioti
  2

• Alloc zero size memory on old model GPU may fail.

  #870 opened a year ago by optman
  0

• Impossible to create a module with a parameter that lacks a const shape?

  #839 opened a year ago by emchristiansen
  1

• Please minimize the requirements for the optimizers

  #840 opened a year ago by emchristiansen
  1

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

  #866 opened a year ago by jackerschott
  2

• ConcatAlong is very slow

  #867 opened a year ago by opfromthestart
  12

• Expose AddDim to public API

  #859 opened a year ago by LeoDog896
  1

• Convenience of mixed tensor scalar operations

  #862 opened a year ago by jackerschott
  1

• Question: How to load checkpoints?

  #861 opened a year ago by rjzak
  1

• Grouped Linear layers

  #857 opened a year ago by opfromthestart
  1

• How to calculate loss to maximize a value

  #850 opened a year ago by opfromthestart
  3

• Differing behavior between Pytorch and dfdx

  #848 opened a year ago by opfromthestart
  6

• impl `Merge` for `NoneTape` for `OwnedTape` and `Arc<Mutex<OwnedTape>>`

  #836 opened a year ago by emchristiansen
  1

• Consider adding a `Tape` type where the tape is optional, e.g. `Option<Arc<Mutex<OwnedTape<_, _>>>>`

  #837 opened a year ago by emchristiansen
  2

• Bug: `as_vec` crashes for empty Tensors

  #844 opened a year ago by emchristiansen
  0

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

  #841 opened a year ago by emchristiansen
  2

• Bug: Incompatible uniqueness assumption causes crashes

  #842 opened a year ago by emchristiansen
  1

• `ArcTape`: Add another `Tape` impl?

  #833 opened a year ago by emchristiansen
  4

• Is it possible to train models with multiple inputs?

  #826 opened a year ago by emchristiansen
  7

• Add Gradient scaler

  #832 opened a year ago by coreylowman
  1

• Why do gradient tapes need to start at the inputs?

  #827 opened a year ago by emchristiansen
  2