Skip to content

v0.5.0
Compare
Choose a tag to compare
@dfalbel dfalbel released this 17 Aug 16:43
v0.5.0
55ee872
This commit was signed with the committer’s verified signature.
dfalbel Daniel Falbel
GPG key ID: 86D67393B1F8D380
Learn about vigilant mode.

Breaking changes

  • Droped support for CUDA 10.1 (#610)
  • torch_manual_seed() now matches PyTorch's behavior so we can more easily compare implementations. Since this is a breaking change we added the torch.old_seed_behavior=TRUE option so users can stick to the old behavior. (#639)
  • Indexing with vectors has a now the same behavior as R indexing, making it easier to understand. Users can still use the old behavior by using torch_index or torch_index_put. (#649)

New features

  • Added support for ScriptModule. Loaded JIT modules now operate as nn_modules. (#593)
  • Added a jit_compile function that allows compiling arbitrary TorchScript code into script function that can be serialized and executed. (#601)
  • Added jit_trace support for nn_module created from R. (#604)
  • Updated LibTorch to version 1.9.0 (#610)
  • Added Linear Algebra functions (#612)
  • Added contrib_sort_vertices to efficiently sort vertices on CUDA. (#619)
  • Allows querying the graph from traced modules. (#623)
  • Added with_detect_anomaly to debug autograd errors. (#628)
  • Implemented traced_module$graph_for() to allow inspecting the optimized jit graph. (#643)
  • Added slc to allow dynamically creating slices when indexing tensors. (#648)

Bug fixes

  • Fixed a bug when using a .getbatch method that didn't return a torch_tensor. (#615)
  • Fixed warning when using %/% caused by a call to deprecated torch_floor_divide (#616)
  • Improved CUDA version auto-detection (#644)

Internal changes

  • Improved R <-> JIT types conversion. (#593)
  • Added Dockerfiles for CUDA 11.1 (#597)
  • A warning is raised when an incompatible dataset is passed to a parallel dataloader. (#626)
  • Additionally to calling gc when CUDA memory is exhausted we now call R_RunPendingFinalizers. This should improve memory usage, because we will now delete tensors earlier. (#654)
  • Fix rchk issues (#667)
Assets 2