All pull requests

Pull requests

• Fix inference layer_norm/rms_norm launching no kernel for fp32 hidden sizes 65-128

  #8827·jayden1711 opened Oct 15, 2026·

  ·
  1

  3

• [CI] Run torch-latest only on AWS

  #8825·tohtana opened Oct 14, 2026Collaborator·

  ·
  8

• Fix routed-expert gradient clipping for AutoEP + AutoTP folding

  #8823·0z5a opened Oct 14, 2026Contributor·

  ·
  4

• Respect BF16 overflow-check opt-out in ZeRO-1/2

  #8818·XFDG opened Oct 13, 2026·

  ·
  1

  6

• Add RVV vectorized FP32 kernel to CPU Adam

  #8817·zhangjian29 opened Oct 13, 2026·

  ·
  8

• Use C++20 for NPU op builders on torch 2.12+

  #8816·VenusTZZ opened Oct 13, 2026Contributor·

  ·
  1

• Add deepspeed.comm.all_reduce_if_parallel for optional parallel groups

  #8815·jinyouzhi opened Oct 13, 2026Contributor·

  ·

• Build CPU optimizer ops without CUDA and fix the CPU Lion SIMD update

  #8813·hwchen2017 opened Oct 13, 2026Contributor·

  ·
  1

  4

• [AutoSP] Align cross entropy targets after an in-model causal label shift

  #8811·jinyouzhi opened Oct 12, 2026Contributor·

  ·
  12

• [checkpoint] Execute affine transfer plans with bounded peer staging

  #8809·0z5a opened Oct 12, 2026Contributor·

  ·
  7

• fix(fp_quantizer): rename local shadowing template param for CUDA 13+ nvcc

  #8808·bheu opened Oct 12, 2026Contributor·

  ·
  4

• Speed up import deepspeed by deferring heavy imports and import-time work

  #8807·hwchen2017 opened Oct 12, 2026Contributor·

  ·
  11

• Add a nightly multi-rank CPU suite

  #8806·delock opened Oct 12, 2026Collaborator·

  ·
  10

• Accept warmup_num_steps=0 in the warmup LR schedulers again

  #8804·vineethsaivs opened Oct 10, 2026Contributor·

  ·

• Warn when the Ulysses SP DataLoader is not sharded across the SP group

  #8802·varun-projects opened Oct 10, 2026·

  ·
  4

• Guard the deprecated tensor model parallel mpu fallbacks

  #8801·varun-projects opened Oct 10, 2026·

  ·
  1

  4

• [CKPT] Plan expert transfers from the live runtime placement

  #8800·jinyouzhi opened Oct 10, 2026Contributor·

  ·
  4

• Add KLXPU (KUNLUNXIN XPU) accelerator

  #8799·GuCarpenter opened Oct 10, 2026·

  ·
  12

• comms_logger: honor in-section enabled and bound per-op timer history

  #8798·hgdytzyy opened Oct 10, 2026·

  ·
  1

  8

• Skip accelerator sync for unprofiled comm ops

  #8795·guptaishaan opened Oct 10, 2026·

  ·
  1

  2

• Average each loss of a tuple in pipeline eval_batch

  #8793·vineethsaivs opened Oct 9, 2026Contributor·

  ·

• Add vocabulary-parallel fused linear cross entropy

  #8792·jinyouzhi opened Oct 9, 2026Contributor·4 tasks·

  ·
  9

• Add vocabulary-parallel logprobs, entropy and KL ops

  #8791·jinyouzhi opened Oct 9, 2026Contributor·

  ·
  15

• Restore the data sampler's own RNG when resuming from a checkpoint

  #8788·Rakshit-gen opened Oct 9, 2026Contributor·

  ·
  1

  6

• perf(rollout): select generation and prefill buckets with a tunable cost model

  #8786·SuperGoodGame opened Oct 9, 2026·

  ·
  1

  27