Issues

• Build Docs upload job: git push -f to gh-pages rejected (cannot lock ref, concurrent update)

  ci-failure
  pytorch/ao#4964 · Filter by author github-actions opened on Sep 30, 2026
  1

• test_triton_mxfp8_dim1_randn[FLOOR-256-128]: exact e8m0 scale mismatch (12/1024) on ROCm MI350

  ci-failuremodule: rocm
  pytorch/ao#4953 · Filter by author github-actions opened on Sep 25, 2026
  1

• RFC: Updated TFMOT-compatible 2:4 for Linear and Conv2d

  pytorch/ao#4949 · Filter by author wwwind opened on Sep 25, 2026
  1

• [README] Dead link in getting started

  pytorch/ao#4942 · Filter by author AbdallahMeebed opened on Sep 24, 2026
  3

• Run-time dependencies aren't declared

  pytorch/ao#4940 · Filter by author yurivict opened on Sep 23, 2026
  1

• mxfp8 training: single-qdata 2D (32x32) weight blocks now viable - cuBLAS 13.6 lifts the TN-only restriction (sm_120)

  pytorch/ao#4938 · Filter by author ultism opened on Sep 23, 2026

• [mxfp8 training] mx_mm AUTO gemm broken with torch>=2.12: F.scaled_mm v2 requires SWIZZLE_32_4_4 scales

  pytorch/ao#4929 · Filter by author ultism opened on Sep 21, 2026
  1

• [Bug] AdamW8bit picks a different state precision after sharding the same DTensor parameter

  bug
  pytorch/ao#4912 · Filter by author gss10282023 opened on Sep 18, 2026

• pt2e observers silently discard the imaginary part of complex tensors

  bug
  pytorch/ao#4910 · Filter by author rohday opened on Sep 17, 2026
  1

• Support input-channel MX quantization for nn.Conv2d weights

  pytorch/ao#4866 · Filter by author catcor01 opened on Sep 4, 2026
  1

• Support nn.Conv2d in the observer-based Static FP8 quantization flow

  pytorch/ao#4865 · Filter by author catcor01 opened on Sep 4, 2026
  1

• test_dtensor.py (float8 TP compile): aten.abs.default NotImplemented on Float8TrainingTensor after removing allow-in-graph autograd wrappers

  ci-failure
  pytorch/ao#4864 · Filter by author github-actions opened on Sep 2, 2026
  33

• Avoid re-registering native Enum pytree constants

  pytorch/ao#4848 · Filter by author malaiwah opened on Aug 29, 2026

• [MXFP8 training] Weight wrapper drops torch.nn.functional.linear bias

  pytorch/ao#4846 · Filter by author GotFusion opened on Aug 29, 2026

• Int8Tensor/Float8Tensor missing aten.view/abs/arithmetic op registrations in 0.18 (quantize fails silently, unquantized)

  pytorch/ao#4845 · Filter by author JWLHS opened on Aug 28, 2026
  5

• Stale MXFP4 training entry in workflow documentation

  pytorch/ao#4841 · Filter by author GotFusion opened on Aug 28, 2026

• [WIP] MXTensor inference weights cannot be sharded with FSDP2

  pytorch/ao#4837 · Filter by author GotFusion opened on Aug 27, 2026

• MXFP8 conversion unnecessarily requires PyTorch nightly for non-DTensor paths

  pytorch/ao#4830 · Filter by author GotFusion opened on Aug 26, 2026

• generate_permute_indices: the GPU path writes past the end of its max_len output buffer

  pytorch/ao#4825 · Filter by author truong-v opened on Aug 25, 2026
  1

• IndexError in _fqn_to_config_handler when multiple parameter FQNs are set to None

  bug
  pytorch/ao#4818 · Filter by author gaurav-redhat opened on Aug 24, 2026

• Bitpacking default pack and unpack dimensions do not round-trip multidimensional tensors

  pytorch/ao#4815 · Filter by author aswanth-07 opened on Aug 23, 2026

• Track deletion of low-usage TorchAO APIs

  module: coretopic: for developerstracker
  pytorch/ao#4797 · Filter by author iamzainhuda opened on Aug 19, 2026

• Float8Tensor zero inputs quantize to NaN

  pytorch/ao#4713 · Filter by author Rajioba1 opened on Aug 6, 2026

• ConvRot — Group-wise Regular Hadamard Rotation for W4A4/W8A8 Diffusion Transformer Quantization

  enhancement
  pytorch/ao#4695 · Filter by author gaurav-redhat opened on Aug 5, 2026
  5

• ZeroPointDomain (class) vs zero_point_domain (variable) typo

  bug
  pytorch/ao#4693 · Filter by author gaurav-redhat opened on Aug 5, 2026