All pull requests

Pull requests

• Honor the dtype argument of torch sum, mean and prod

  #2927·devteamaegis opened Nov 14, 2026·

  ·

• Fix torch cosine_similarity: clamp each norm, avoid fp16 overflow

  #2926·0z5a opened Nov 13, 2026·

  ·
  1

• Support torch.eye with a runtime size

  #2924·sohailshk opened Nov 13, 2026·

  ·

• Fix torch.argmax and torch.argmin without dim

  #2923·sohailshk opened Nov 13, 2026·

  ·

• Add torch.lerp to the PyTorch converter

  #2921·Ishkirat-Singh opened Nov 11, 2026·

  ·

• Select the item when indexing a shape list in scripted torch models

  #2919·sandeepvijayarao09 opened Nov 11, 2026·

  ·
  1

• Fix load_from_path docstrings and examples

  #2918·electron-rare opened Nov 10, 2026·

  ·

• Fix _python_version parsing for pre-release Python versions

  #2917·laasyapriya2007 opened Nov 7, 2026·

  ·
  1

• Use sys.version_info in _python_version

  #2916·Fire162 opened Nov 7, 2026·

  ·
  1

• Give each repeated PyTorch graph output its own Var

  #2915·sandeepvijayarao09 opened Nov 6, 2026·

  ·
  1

• Support InstanceNorm3d in the PyTorch converter

  #2914·sandeepvijayarao09 opened Nov 5, 2026·

  ·
  1

• fuse_linear_bias: fold a scalar add/sub into the linear bias

  #2911·msluszniak opened Oct 23, 2026·

  ·

• Honor explicit dtype before computing torch.cumprod

  #2910·HowieYin opened Oct 22, 2026·

  ·

• Preserve int32 data when its 16-bit range is unknown

  #2909·HowieYin opened Oct 22, 2026·

  ·

• Match PyTorch clamp bound ordering and NaN propagation

  #2908·HowieYin opened Oct 21, 2026·

  ·

• convert _upsample_bilinear2d_aa with the antialias filter

  #2907·HuzaifaChaudary opened Oct 21, 2026·

  ·

• Skip float32 weights in post-training quantization

  #2904·vera-rubin opened Oct 20, 2026·

  ·

• Add torch.unsafe_chunk to the PyTorch converter (LSTMCell, GRUCell)

  #2903·GodfredArhinful opened Oct 19, 2026·

  ·
  1

• Fix per-block ConvTranspose quantization validation

  #2902·vera-rubin opened Oct 18, 2026·

  ·

• Fix shared constant mutation in convolution bias fusion

  #2901·kadyrbekovhamit-cyber opened Oct 16, 2026·

  ·

• Fix fp16 vector palettization LUT indices

  #2900·vera-rubin opened Oct 16, 2026·

  ·

• convert _native_multi_head_attention , the fast path of nn.MultiheadAttention

  #2899·HuzaifaChaudary opened Oct 15, 2026·

  ·
  1

• [Docs] Add Simplified Chinese translation (README.zh.md)

  #2898·JasonYeYuhe opened Oct 13, 2026·

  ·

• Add torch upsample_trilinear3d and upsample_nearest3d lowering

  #2897·Cassian433 opened Oct 12, 2026·

  ·
  1

• Add torch.hypot to PyTorch converter

  #2896·Manjarly opened Oct 12, 2026·

  ·