All pull requests

Pull requests

• [CI] Restore Modal-first torch-latest routing

  #8849·tohtana opened Oct 16, 2026Collaborator·

  ·
  1

• AutoEP + AutoTP folding: Universal Checkpointing under ZeRO-3

  #8848·pengdurice opened Oct 16, 2026Contributor·

  ·

• AutoEP: support combine_impl="fused_weighted_sum" with AutoEP+AutoTP folding

  #8847·pengdurice opened Oct 16, 2026Contributor·

  ·

• Release process-group caches before CPU/Gloo worker shutdown

  #8846·tohtana opened Oct 16, 2026Collaborator·

  ·
  6

• AutoTP: send CPU copies in the first-batch consistency check

  #8845·pengdurice opened Oct 16, 2026Contributor·

  ·

• Universal Checkpointing: read AutoEP ZeRO-3 optimizer files one at a time

  #8844·pengdurice opened Oct 16, 2026Contributor·

  ·

• Keep the parameters a model names for fp32 in fp32 under ZeRO-3

  #8843·pengdurice opened Oct 16, 2026Contributor·

  ·

• ZeRO-3: reuse the weight partition as the master copy when it already has the master dtype

  #8842·pengdurice opened Oct 16, 2026Contributor·

  ·

• Peng add cpu adam fused step v1

  #8841·pengdurice opened Oct 16, 2026Contributor·

  ·

• ZeRO-3: copy offloaded master weights back to the device through pinned buffer

  #8840·pengdurice opened Oct 16, 2026Contributor·

  ·

• [OPSD] Request last-position logits only in HybridEngine rollout prefill

  #8839·jinyouzhi opened Oct 16, 2026Contributor·

  ·

• Add Triton kernels for vocab-parallel logprobs, entropy and KL

  #8838·jinyouzhi opened Oct 16, 2026Contributor·

  ·

• Fix lost TP all-reduce results for non-contiguous tensors

  #8837·jinyouzhi opened Oct 16, 2026Contributor·

  ·
  4

• Skip the accelerator-only CPU optimizer tests on the cpu accelerator

  #8836·delock opened Oct 16, 2026Collaborator·

  ·
  1

• Fix TestLionConfigs on the CPU accelerator and run it on every runner

  #8833·hwchen2017 opened Oct 16, 2026Contributor·

  ·
  1

• Fix out-of-bounds row access in inference layer_norm/rms_norm kernels for hidden sizes <= 128

  #8830·jayden1711 opened Oct 15, 2026·

  ·
  1

  6

• AutoEP + AutoTP folding: allow CPU optimizer offload under ZeRO-3

  #8829·0z5a opened Oct 15, 2026Contributor·

  ·
  1

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

  #8827·jayden1711 opened Oct 15, 2026·

  ·
  1

  4

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

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

  ·
  4

• Add RVV vectorized FP32 kernel to CPU Adam

  #8817·zhangjian29 opened Oct 13, 2026·

  ·
  8

• Add deepspeed.comm.pg_utils.is_parallel for optional parallel groups

  #8815·jinyouzhi opened Oct 13, 2026Contributor·

  ·
  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

• Accept warmup_num_steps=0 in the warmup LR schedulers again

  #8804·vineethsaivs opened Oct 10, 2026Contributor·

  ·