All pull requests

Pull requests

• [Test] add deterministic reducescatter in sharding_optimizer

  #79855·Waynezee opened Oct 10, 2026Contributor·

  ·
  4

• [compat] Support torch-style broadcast_to, float8 kernels and distributed.all_to_all_single

  #79854·Difers opened Oct 10, 2026Contributor·

  ·

• [release/3.4] add slice_concat for hyperball

  #79851·xxyux opened Oct 9, 2026Member·

  ·
  3

• add slice_concat for hyperball

  #79850·xxyux opened Oct 9, 2026Member·

  ·
  4

• update vmm

  #79849·liuruyan opened Oct 9, 2026Contributor·

  ·
  4

• [CI] Use Paddle-Review-Bot as the only bot gate approver (release/3.4)

  contributor
  #79848·Paddle-Review-Bot opened Oct 9, 2026Contributor·

  ·
  1

• [BugFix] Reject negative padding in pooling to match conv

  contributor
  #79846·MincongZhou opened Oct 8, 2026·

  ·

• [Prototype] Support per-group inner optimizers in Muon sharding

  #79844·xxyux opened Oct 8, 2026Member·

  ·
  5

• [Operator Mechanism] Reject illegal 0-size expand_as / split configs in InferMeta

  #79843·feixi139 opened Oct 8, 2026Contributor·

  ·
  4

• [Cherry-Pick][Operator Mechanism]Add K-dim consistency check to matmul_grad InferMeta

  #79839·feixi139 opened Oct 8, 2026Contributor·

  ·
  2

• [BugFix] Isolate Gloo rendezvous keys for each process group

  contributor
  #79838·cyanseek opened Oct 2, 2026·

  ·
  1

• [BugFix] Keep Gloo P2P operations from advancing collective tags

  contributor
  #79837·cyanseek opened Oct 2, 2026·

  ·
  1

• [API Alignment] Add addcmul and addcmul_ support

  contributor
  #79836·Manfredss opened Oct 2, 2026Contributor·

  ·
  5

• test mac-1

  #79834·Ray961123 opened Sep 29, 2026Collaborator·

  ·
  4

• [Operator Mechanism] Fill moe_permute output buffers that no token lands on

  #79832·feixi139 opened Sep 29, 2026Contributor·

  ·

• [Operator Mechanism] Fix O(n^2) python stack accumulation in SetPythonStack

  #79831·feixi139 opened Sep 29, 2026Contributor·

  ·

• [Cherry-Pick][Operator Mechanism] Fix silent precision loss in GPU eigh/eigvalsh by restoring the syevjBatched n<=32 gate

  #79830·feixi139 opened Sep 29, 2026Contributor·

  ·
  3

• [Operator Mechanism] Fix silent precision loss in GPU eigh/eigvalsh by restoring the syevjBatched n<=32 gate

  #79829·feixi139 opened Sep 29, 2026Contributor·

  ·
  3

• [API Compatibility No.71] Add out parameter support for histogram -part

  contributor
  #79825·zhengguangzhuo opened Sep 28, 2026·

  ·
  2

• [Typing] Adapt stub generation to pybind11-stubgen 3.0

  contributor
  #79821·SigureWang opened Sep 28, 2026Contributor·

  ·

• [Typing] Adapt stub generation to pybind11-stubgen 3.0

  contributor
  #79820·SigureWang opened Sep 28, 2026Contributor·

  ·
  2

• [Cherry-Pick][Operator Mechanism] Fix topk returning NaN for an empty reduction axis

  #79818·feixi139 opened Sep 28, 2026Contributor·

  ·
  1

• using phi::CastOpTransformFunctor to cast float16/bfloat16 to complex

  contributor
  #79816·Malone-AI opened Sep 28, 2026·

  ·
  14

• [Fsdp] support fp8 and fused linear

  #79813·zhengshengning opened Sep 24, 2026Contributor·

  ·
  4

• Stabilize Code Sync submodule packaging in Clone commit and package job

  contributor
  #79802·XieYunshen with CopilotAI opened Sep 22, 2026·

  ·
  2