All pull requests

Pull requests

• fix bug for cache file

  #2133·YifanHe-ailab opened Sep 30, 2026Collaborator·

  ·

• [WIP] Controller on datapacker

  #2132·PengchengShi00 opened Sep 29, 2026Collaborator·

  ·

• [Feature] Bind NPU training ranks to NUMA-local CPU slices (XTUNER_NPU_CPU_BIND)

  #2131·longboat2010 opened Sep 27, 2026·

  ·

• [Feature] Route FSDP2 unshard to the compute stream (XTUNER_FSDP_UNSHARD_ASYNC_OP)

  #2130·longboat2010 opened Sep 27, 2026·

  ·

• [Feature] NPU liger fused-linear-cross-entropy loss (XTUNER_LOSS_MODE=liger)

  #2129·longboat2010 opened Sep 27, 2026·

  ·

• [Feature] Fused SwiGLU Triton kernels for MLP activation (XTUNER_NPU_FUSED_SWIGLU)

  #2128·longboat2010 opened Sep 26, 2026·

  ·

• fix(datasets): build the DP all_reduce tensor on the mesh device instead of assuming CUDA

  #2127·li-lizhe opened Sep 25, 2026·

  ·
  1

• [Feature] Keep Muon and AdamW optimizer state on host

  #2125·jayhenry opened Sep 24, 2026Collaborator·

  ·
  4

• [Fix] Restore Muon optimizer states on DCP resume

  #2124·jayhenry opened Sep 24, 2026Collaborator·

  ·
  3

• [Fix] Use at least num_sms DeepEP QPs so internode EP16 passes the IBGDA check

  #2123·jayhenry opened Sep 24, 2026Collaborator·

  ·

• [ ci ] add interns2-preivew cases on npu

  #2121·kkscilife opened Sep 24, 2026Collaborator·

  ·

• [Feature] NPU AscendC op stack for GatedDeltaNet (causal conv1d / chunk delta rule / fused gated norm)

  #2120·longboat2010 opened Sep 24, 2026·

  ·

• [Feature] NPU fused zero-centered RMSNorm (XTUNER_NPU_FUSED_ZC_NORM)

  #2119·longboat2010 opened Sep 24, 2026·

  ·

• [ ci ] add qwen3.5 cases on npu

  #2117·kkscilife opened Sep 24, 2026Collaborator·

  ·

• [Fix] Initialize zero-centered RMSNorm weight to zeros

  #2116·YanhuiDua opened Sep 24, 2026Collaborator·

  ·

• [Feature] Add offline token and packing statistics for text datasets

  #2115·bingshao333 opened Sep 24, 2026·

  ·
  1

• [fix] IPC barrier

  #2114·PengchengShi00 opened Sep 24, 2026Collaborator·

  ·
  1

• [Fix]: stabilize Qwen3.5 Dense gradients across sequence parallel sizes

  #2113·JT-Ushio opened Sep 24, 2026Contributor·

  ·
  6

• [Feature] Add GLM-5.3-Flash F6 core: text model + MTP + compose model

  #2111·jayhenry opened Sep 23, 2026Collaborator·7/7·

  ·

• [Feature] Add GLM-5.3-Flash F2: vision tower + projector (eager)

  #2110·jayhenry opened Sep 23, 2026Collaborator·6/7·

  ·

• [Feature] Add GLM-5.3-Flash F1: VL data preprocessing pipeline

  #2109·jayhenry opened Sep 23, 2026Collaborator·5/7·

  ·

• [Feature] Add GLM-5.3-Flash F5: NoPE DSA + KPool indexer + clamped SwiGLU

  #2108·jayhenry opened Sep 23, 2026Collaborator·4/7·

  ·

• [Feature] Add GLM-5.3-Flash F4: mHC four-stream residual

  #2107·jayhenry opened Sep 23, 2026Collaborator·3/7·

  ·

• [Feature] Add GLM-5.3-Flash F3: Kimi Delta Attention (KDA)

  #2106·jayhenry opened Sep 23, 2026Collaborator·2/7·

  ·

• [Feature] Add GLM-5.3-Flash F0: 25B cropped reference checkpoint builder

  #2105·jayhenry opened Sep 23, 2026Collaborator·1/7·

  ·