-
Notifications
You must be signed in to change notification settings - Fork 51
Pull requests: modelscope/twinkle
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
fix: weight PPO token-mean micro-batches by valid tokens
#292
opened Sep 20, 2026 by
taking-lying-flat
Loading…
1 of 4 tasks
fix: detach probability weights in DFT loss
#291
opened Sep 20, 2026 by
taking-lying-flat
Loading…
1 of 4 tasks
fix: avoid duplicate gradient norm reduction in pure DDP
#290
opened Sep 20, 2026 by
taking-lying-flat
Loading…
1 of 4 tasks
feat(server): execution guarantees, async lifecycle, and strict wire contracts
#286
opened Sep 15, 2026 by
Yunnglin
Collaborator
Loading…
fix(rl): reduce Megatron memory usage in selective_log_softmax, causal mask, and discarded logits
#283
opened Sep 14, 2026 by
Wings-Of-Disaster
Collaborator
Loading…
1 of 4 tasks
feat(rl): add SAO synchronous training baseline with skip-observation GAE, DIS policy loss and value critic ```
#269
opened Aug 19, 2026 by
xxyyrr598
Contributor
Loading…
Fix/multi lora ep memory efficient init
#265
opened Aug 14, 2026 by
meichangsu1
Collaborator
•
Draft
4 tasks
ProTip!
What’s not been updated in a month: updated:<2026-08-21.