v0.17.0

Breaking changes

• Updated to LibTorch 2.8.0 (#1419).
• torch_triu_indices() and torch_tril_indices() now return 1-based indexes (#1382).
• $indices() now returns 1-based indexes (#1382).

New features

• Added torch_sitrep() for torch installation situation report (#1415, @cregouby).
• Added %*% operator for torch tensors (#1379, @caio-hamamura).
• Added head() and tail() methods for torch tensors (#1388).
• Exported torch_scaled_dot_product_attention() (#1404).
• Exported torch_sparse_sampled_addmm() (#1427).
• Exported torch_ldexp() (#1407).
• Integrated cudatoolkit R package for CUDA library loading (#1422).
• Improved as_array() support for sparse tensors (#1387).

Bug fixes

• Fixed sparse CSR tensor printing, to_sparse() crash, and as_array() detection (#1431).
• Fixed dataset()/nn_module() crash when initialize parameter is named d (#1424, @chandraveer-singh).
• Fixed bfloat16 not being recognized as a floating point dtype (#1408).
• Fixed handling of R longjump exceptions in autograd and tracing callbacks (#1406).
• Fixed redundant lantern.h include causing macro redefinition warning (#1430).
• Excluded glibc libraries from dependency bundling (#1397, @TroyHernandez).
• Replaced non-API entry point Rf_findVarInFrame with R_getVarEx for R 4.6 compatibility (#1421).
• Added French translations for 5 missing messages (#1385, @cregouby).

v0.16.3

Increment version number to 0.16.3

v0.16.2

More fixes related to CRAN submissions

v0.16.1

• Fixed issue with Windows CRAN checks. (#1366)

v0.16.0

• Support for CUDA 11.8 was dropped on Windows. (#1342)
• Dropped support for CUDA 12.4 (#1348)
• Added Support for CUDA 12.6 and 12.8 (#1348)
• Updated to LibTorch 2.7.1 (#1348)
• Fixed a bug causing repeat_interleave to not work properly on CUDA (#1353)

v0.15.1

Increment version number to 0.15.1

v0.15.0

• add ROC AUM loss with nn_aum_loss() (#1310 @cregouby)
• Add translation of install messages in French (@cregouby #1317)
• fix installation checks for r2u users ((@cregouby #1317)

v0.14.2

• Fixed a regression causing torch to crash on Windows if used within RStudio.

v0.14.1

Bug fixes:

• Fixed issue when compiling on non-glibc Linux distributions that don't implement RTLD_DEEPBIND (#1268)

v0.14.0

Breaking changes

• Updated to LibTorch v2.5.1 (#1204) -- potentially breaking change!

New features

• Feature: Faster optimizers (optim_ignite_<name>()) are available: Adam, AdamW, Adagrad, RMSprop,SGD.
  These can be used as drop-in replacements for optim_<name> but are considerably
  faster as they wrap the LibTorch implementation of the optimizer.
  The biggest speed differences can be observed for complex optimizers such as AdamW.

Bug fixes

• torch_iinfo() now support all integer dtypes (#1190 @cregouby)
• Fixed float key_padding_mask in nnf_multi_head_attention_forward() (#1205)
• Fix french translation (#1176 @cregouby)
• Trace jitted modules now respect 'train' and 'eval' mode (#1211)

• Fix: Avoid name clashes between multiple calls to jit_trace (#1246)