All pull requests

Pull requests

• [Fsdp] support fp8 and fused linear

  #79813·zhengshengning opened Sep 24, 2026Contributor·

  ·
  3

• [Cherry-Pick][compat] Resolve a scoped spec once instead of re-entering it

  #79812·LiYuRio opened Sep 23, 2026Contributor·

  ·
  1

• [Cherry-Pick] Support iteration space beyond INT32_MAX in GPU index_elementwise_put kernels(#79810)

  #79811·omoYang opened Sep 22, 2026Contributor·

  ·
  1

• [FlexCheckpoint] Avoid the full metadata load in the local-resume check

  #79807·DanielSun11 opened Sep 22, 2026Contributor·

  ·
  1

• [Cherry-Pick] Fix new[]/delete mismatch in AttributeManager destructor

  #79806·omoYang opened Sep 22, 2026Contributor·

  ·
  1

• [Execute Infrastructure] Fix new[]/delete mismatch in AttributeManager destructor

  #79805·omoYang opened Sep 22, 2026Contributor·

  ·
  2

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

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

  ·
  2

  

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

  #79800·feixi139 opened Sep 22, 2026Contributor·

  ·
  4

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

  #79799·feixi139 opened Sep 22, 2026Contributor·

  ·
  4

• [Operator Mechanism] Pre-reduce broadcast axes in strided view backward

  #79796·feixi139 opened Sep 21, 2026Contributor·

  ·
  5

• [release/3.4] Translate Chinese comments to English in muon optimizer test

  #79795·xxyux opened Sep 21, 2026Contributor·

  ·

• [Cherry-Pick][Performance] Add num_workers to paddle.load for parallel payload reading (#79786)

  #79787·DanielSun11 opened Sep 17, 2026Contributor·

  ·
  10

• [Cherry-Pick] Add deterministic path for take_along_axis_grad cuda kernel

  skip-ci: api-benchmarkskip-ci: slice
  #79785·omoYang opened Sep 17, 2026Contributor·

  ·
  11

• [FSDP] FSDP Support Muon and CSA

  #79781·Xing-lil opened Sep 16, 2026Contributor·

  ·
  7

• [Cherry-Pick] Align cosine_similarity with torch2.12

  #79779·omoYang opened Sep 15, 2026Contributor·

  ·
  15

• [Cherry-Pick] optimize cummax/cummin GPU scan kernels

  #79778·omoYang opened Sep 15, 2026Contributor·

  ·
  8

• Add compatibility for CUB/CCCL 3.0+ API

  #79777·omoYang opened Sep 15, 2026Contributor·

  ·
  17

• [Cherry-Pick] Fix 0-D index bug and add deterministic path in gather grad kernel

  #79773·omoYang opened Sep 15, 2026Contributor·

  ·
  8

• [Test] Un-shadow duplicated test classes in test_lu_op.py and test_tensordot.py

  contributor
  #79767·Anai-Guo opened Sep 12, 2026·

  ·
  1

• [XPU] Remove duplicated activation op-test wrappers that shadow sqrt and silu tests

  contributor
  #79766·Anai-Guo opened Sep 12, 2026·

  ·
  1

• [AP] Add conv2d + epilogue fusion pattern.- #79532

  #79765·Xreki opened Sep 11, 2026Contributor·

  ·
  7

• [Test] Rename AI-edited tests to coverage tests

  skip-ci: api-benchmarkskip-ci: slice
  #79761·liuhao2638 opened Sep 10, 2026Contributor·

  ·
  3

• [Bug Fix] Avoid overflow in CPU SiLU backward

  contributor
  #79760·MrCapricornLiu opened Sep 10, 2026·

  ·
  1

• [Bug Fix] Gate the fp8 fast transpose on the base pointer alignment

  #79757·DanielSun11 opened Sep 9, 2026Contributor·

  ·
  3

• [Operator Mechanism] Add deterministic path for cummax, cummin and cumsum

  #79755·omoYang opened Sep 9, 2026Contributor·

  ·
  15