Skip to content

Pull requests: THUDM/slime

New pull request New
138 Open 1,425 Closed
138 Open 1,425 Closed
Author
Filter by author
Loading
Label
Filter by label
Loading
Use alt + click/return to exclude labels
or ⇧ + click/return for logical OR
Projects
Filter by project
Loading
Milestones
Filter by milestone
Loading
Reviews
Filter by reviews
No reviews Review required Approved review Changes requested
Assignee
Filter by who’s assigned
Assigned to nobody Loading
Sort
Sort by
Newest Oldest Most commented Least commented Recently updated Least recently updated Best match
Most reactions
👍 👎 😄 🎉 😕 ❤️ 🚀 👀

Pull requests list

[draft] Add support for multi teacher OPD and memory-efficient topk level OPD
#2033 opened Jun 8, 2026 by hhnqqq • Draft
Support OPD when teacher tokenization differs
#2032 opened Jun 8, 2026 by hhnqqq Loading…
1
examples: add CISPO custom loss
#2026 opened Jun 6, 2026 by kekmodel Contributor Loading…
fix(rollout): drain generation before offload memory release
#2015 opened Jun 4, 2026 by EazyReal Loading…
fix(rollout): apply sample filter in rollout manager
#2014 opened Jun 4, 2026 by EazyReal Loading…
fix(colocate): derive num_gpus_per_node from actor_num_gpus_per_node
#2012 opened Jun 3, 2026 by aoshen02 Contributor Loading…
perf(ppo): reduce log-prob + entropy cross-entropy peak memory
#2011 opened Jun 2, 2026 by Mantissagithub Loading…
1
perf(megatron-loss): scale logits per-chunk to avoid OOM
#2010 opened Jun 2, 2026 by Yangruipis Contributor Loading…
perf(rollout): pack loss_masks as np.int8 at the ray.put boundary
#2006 opened Jun 2, 2026 by Chasing1020 Contributor Loading…
3
[Draft] Refactor trajectory manager
#2005 opened Jun 2, 2026 by jingshenghang Collaborator • Draft
19
fix(gpt-oss): emit fused raw expert tensors for SGLang
#2004 opened Jun 1, 2026 by Jiang020609 Contributor Loading…
1
fix: clamp max_new_tokens on retry to prevent response_length overflow
#2003 opened Jun 1, 2026 by YaoweiFan Loading…
fix(megatron): honor --[no-]gradient-accumulation-fusion on the megatron.bridge provider
#1999 opened Jun 1, 2026 by aoshen02 Contributor Loading…
fix(rollout): support non extra gpu placement when using rollout-external mode
#1997 opened May 30, 2026 by shinytang6 Loading…
1
fix(logging): partition raw rewards for correct samples
#1996 opened May 30, 2026 by Jiang020609 Contributor Loading…
1
[sft] rebuild the sft loss mask generator and add ci
#1994 opened May 30, 2026 by zhuzilin Contributor Loading…
Add timeout configuration for on policy distillation HTTP session.
#1970 opened May 28, 2026 by qqwqqw689 Contributor Loading…
Optimize CP sequence KL communication for GSPO/OPSM
#1948 opened May 26, 2026 by zzdeae86 Loading…
2
fix:TorchMemorySaver observes invalid LD_PRELOAD. when add --disable-weights-backuper
#1937 opened May 22, 2026 by zyfzjsc988 Loading…
1
Add Trackio rollout trace logging
#1935 opened May 21, 2026 by abidlabs Loading…
feat: add SFT entropy logging and validation loss monitoring
#1925 opened May 19, 2026 by none0663 Contributor Loading…
fix(debug): auto-append rollout_id/rank in save_debug_train_data path template
#1922 opened May 19, 2026 by wlf-darkmatter Loading…
Fix RolloutManager reward normalization for uneven rollout groups
#1918 opened May 18, 2026 by haoyang9804 Loading…
fix: make OPSM reject whole off-policy sequences
#1917 opened May 18, 2026 by haoyang9804 Loading…
2
feat: add --max-checkpoint-count to limit saved checkpoints
#1914 opened May 16, 2026 by JIANG54864 Loading…
ProTip! Adding no:label will show everything without a label.