Issues

• FSDP all-gather autobucketing can apply stale bucket plans after graph mutations

  bugpython
  meta-pytorch/autoparallel#534 · Filter by author AlbedoWang opened on Sep 23, 2026

• Support token-major variable-length context parallelism

  meta-pytorch/autoparallel#532 · Filter by author AlbedoWang opened on Sep 11, 2026

• local_map silently assigns incorrect placements to parameter gradients under AutoParallel

  meta-pytorch/autoparallel#515 · Filter by author yanboliang opened on Jul 9, 2026

• ShardingOptimizer.save()/load() resurrects constraints removed via remove_constraints()

  meta-pytorch/autoparallel#476 · Filter by author fmassa opened on May 26, 2026

• AutoParallel fails on HF transformers Llama-3 / Qwen-2.5: AssertionError in scaled_dot_product_cudnn_attention strategy

  meta-pytorch/autoparallel#456 · Filter by author apbose opened on May 12, 2026

• Move AutoParallel activation checkpointing out of the compile backend

  meta-pytorch/autoparallel#447 · Filter by author sanketpurandare opened on May 5, 2026

• Unifying operator.getitem with the standard code path

  enhancement
  meta-pytorch/autoparallel#388 · Filter by author fmassa opened on Mar 22, 2026

• run_with_rng_state + checkpointing interactions

  meta-pytorch/autoparallel#273 · Filter by author xmfan opened on Dec 8, 2025

• Compile-based Pipeline Parallelism

  meta-pytorch/autoparallel#199 · Filter by author fmassa opened on Oct 9, 2025
  

• Llama3 doesn't work when there's only one GPU

  meta-pytorch/autoparallel#73 · Filter by author ezyang opened on Aug 4, 2025
  1

• Integrate memory tools into default examples

  meta-pytorch/autoparallel#55 · Filter by author ezyang opened on Jul 28, 2025