Commits

Branch selector

main

User selector

All users

All time

Commit history

Commits on Jun 8, 2026

• [veomni] fix: pass mixed precision config to veomni (#6644)

  Show description for 9add650
  

  Luosuu

  and

  codex

  authored
  9add650

  Copy full SHA for 9add650

Commits on Jun 7, 2026

• [rollout] fix: avoid arbitrary code execution in Qwen3 tool parser (#6542)

  Show description for 2084d8f
  

  fuergaosi233

  and

  claude

  authored
  2084d8f

  Copy full SHA for 2084d8f

Commits on Jun 6, 2026

• [ci] fix: Modify the log path of npu nightly ci (#6600)

  Show description for 4ff9ca5
  daikang6

  authored
  4ff9ca5

  Copy full SHA for 4ff9ca5

• [fsdp] feat: add Qwen3.5-4B on-policy distillation FSDP script (#6638)

  Show description for 7757fb0
  cls1206

  authored
  7757fb0

  Copy full SHA for 7757fb0

• [ci] fix: remove uv usage from A2 vlm_rl_job, use pip instead (#6618)

  Show description for 68930b6
  

  3 peopleauthored
  68930b6

  Copy full SHA for 68930b6

Commits on Jun 5, 2026

• [megatron] fix: clamp num_tokens=0 in MTP loss & add normalized scale for MTP per token loss (#6464)

  Show description for 8911f24
  arvyanh

  authored
  8911f24

  Copy full SHA for 8911f24

• [misc] fix: revert run_qwen3_8b_fsdp_npu.sh (#6601)

  Show description for c131c70
  hustmf

  authored
  c131c70

  Copy full SHA for c131c70

• [doc] chore: update amd rocm index (#6622)

  Show description for 189663a
  wuxibin89

  authored
  189663a

  Copy full SHA for 189663a

• [doc] chore: Update index.rst (#6617)

  Show description for 6209469
  hustmf

  authored
  6209469

  Copy full SHA for 6209469

• [fsdp] fix: do not manually move model to GPU (#6604)

  Show description for a539474
  

  pavelgein

  and

  gemini-code-assist[bot]

  authored
  a539474

  Copy full SHA for a539474

• [rollout] chore: change default to calculate_log_probs=True (#6616)

  Show description for 614ffc0
  wuxibin89

  authored
  614ffc0

  Copy full SHA for 614ffc0

• [data] fix: default image_patch_size to processor's real patch_size in filter_overlong_prompts (#6595)

  Show description for f1478d2
  

  zengxingchen

  and

  gemini-code-assist[bot]

  authored
  f1478d2

  Copy full SHA for f1478d2

• [veomni] feat: reduce extra memory from O(E) to O(E/ep_size) during weight refit (#6612)

  Show description for c2d93fc
  wuxibin89

  authored
  c2d93fc

  Copy full SHA for c2d93fc

• [perf] feat: support partial-token window profiling for rollout in verl (#5760)

  Show description for 381ed24
  mengchengTang

  authored
  381ed24

  Copy full SHA for 381ed24

Commits on Jun 4, 2026

• [tool] fix: update npu profiling test scrip (#6606)

  Show description for d5fc06d
  yyyy2000

  authored
  d5fc06d

  Copy full SHA for d5fc06d

• [hardware] feat: add platform abstraction layer and plugin-based engine override system (#6086)

  Show description for 7a457c9
  

  3 peopleauthored
  7a457c9

  Copy full SHA for 7a457c9

• [vllm, megatron] fix: mxfp8 trainning support on Ascend NPU (#6562)

  Show description for 0ce9478
  

  quancs

  and

  gemini-code-assist[bot]

  authored
  0ce9478

  Copy full SHA for 0ce9478

• [tool, rollout] feat: Use SkipManager to uniformly manage the skipping schemes. (#6097)

  Show description for 266fdce
  

  3 peopleauthored
  266fdce

  Copy full SHA for 266fdce

• [ci] fix: chang sglang8.5.0 a3 (#6586)

  Show description for f9e5502
  daikang6

  authored
  f9e5502

  Copy full SHA for f9e5502

Commits on Jun 3, 2026

• [doc] feat: add Qwen3.5-122B NPU guide link (#6588)

  Show description for 74cebc5
  zjchenn

  authored
  74cebc5

  Copy full SHA for 74cebc5

• [model] feat: add Qwen3.5-122B Ascend support (#6582)

  Show description for 81fbf5e
  

  zjchenn

  and

  zhouhengan1211

  authored
  81fbf5e

  Copy full SHA for 81fbf5e

• [ci] fix: stablize main CIs (#6568)

  Show description for e8aff34
  ETOgaosion

  authored
  e8aff34

  Copy full SHA for e8aff34

• [docker] fix: align stable sglang image for GB200 (#6581)

  Show description for cb82110
  

  kaixih

  and

  codex

  authored
  cb82110

  Copy full SHA for cb82110

Commits on Jun 2, 2026

• [doc] chore: Update ascend_sglang_best_practices.rst (#6573)

  Show description for 0c36dee
  hustmf

  authored
  0c36dee

  Copy full SHA for 0c36dee

• [megatron, ckpt] fix: handle None param_data in get_megatron_module_device when use_distributed_optimizer=False (#5782)

  Show description for c34c001
  khazic

  authored
  c34c001

  Copy full SHA for c34c001

• [ci] chore: migrate pip→uv in all Ascend CI workflows (#6535)

  Show description for c64b94b
  

  3 peopleauthored
  c64b94b

  Copy full SHA for c64b94b

• [megatron, ckpt] refactor: optimize mcore ckpt manager impl, checkpoint yaml config APIs (#6014)

  Show description for 22b04f9
  ETOgaosion

  authored
  22b04f9

  Copy full SHA for 22b04f9

• [vllm] fix: reset all caches after weight updates (#6522)

  Show description for accb1a0
  s-isaev

  authored
  accb1a0

  Copy full SHA for accb1a0

• [reward] fix: release worker count when NaiveRouter request fails after retries (#6544)

  Show description for 11ecc15
  

  fuergaosi233

  and

  claude

  authored
  11ecc15

  Copy full SHA for 11ecc15

• [ci] chore: continue to replace the qwen25 model with the qwen3 model (#6554)

  Show description for d880ed3
  daikang6

  authored
  d880ed3

  Copy full SHA for d880ed3

Commits on Jun 1, 2026

• [megatron] fix: Fix GPU memory leak in ref model offload by explicitly releasing storage (#6447)

  Show description for 0a1c5b6
  ZLiao097

  authored
  0a1c5b6

  Copy full SHA for 0a1c5b6

• [ci,veomni] fix: make router-replay install test independent of veomn… (#6557)

  Show description for 738fcef
  hjshi84

  authored
  738fcef

  Copy full SHA for 738fcef

• [misc] chore: bump version to 0.9.0.dev (#6553)

  Show description for 8f5e161
  wuxibin89

  authored
  8f5e161

  Copy full SHA for 8f5e161

• [ci] fix: FSDP actor/critic ci fail (#6550)

  Show description for 7aed6b2
  wuxibin89

  authored
  7aed6b2

  Copy full SHA for 7aed6b2

• [model] fix: support trl>=0.29 AutoModelForCausalLMWithValueHead import (#6539)

  Show description for 3666ee6
  

  wenzhaoabc

  and

  gemini-code-assist[bot]

  authored
  3666ee6

  Copy full SHA for 3666ee6