-
Notifications
You must be signed in to change notification settings - Fork 1.7k
Pull requests: modelscope/ms-swift
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
fix(template): create Janus generation tensors on the input device instead of .cuda()
#10230
opened Sep 22, 2026 by
li-lizhe
Loading…
fix(model): preserve seq-cls metadata across save round trips
#10228
opened Sep 22, 2026 by
Excelius-Wang
Contributor
Loading…
1 task done
[bugfix] skip unobserved GRPO task reward statistics
#10225
opened Sep 22, 2026 by
aprylewu
Contributor
Loading…
1 of 4 tasks
feat: add Moore Threads MUSA backend support via torchada
#10224
opened Sep 21, 2026 by
caizhi-mt
Loading…
1 of 4 tasks
fix(infer): align TransformersEngine with doc defaults by propagating…
invalid
This doesn't seem right
#10222
opened Sep 21, 2026 by
DreamBinary
Loading…
Fix/tool user followup root
postpone
We'll revisit this PR once the refactoring work is finished
#10201
opened Sep 19, 2026 by
tastelikefeet
Collaborator
•
Draft
1 of 4 tasks
fix(rlhf): percent-encode the IPv6 zone id in rollout base URLs
invalid
This doesn't seem right
pending
#10200
opened Sep 19, 2026 by
Redemption-ZTX
Loading…
fix(template): preserve native user follow-ups after tool results
#10198
opened Sep 18, 2026 by
Excelius-Wang
Contributor
Loading…
1 of 4 tasks
[Feat] [FSDP2] [SFT] [ExpertParallel] [DeepSeekV4] fsdp2 dpsk v4 flash 0731 support on only 1Node with 8*H200
postpone
We'll revisit this PR once the refactoring work is finished
#10197
opened Sep 18, 2026 by
mugglewei97
Loading…
feat(sft): support logits selection with Ulysses sequence parallelism
pending
#10191
opened Sep 17, 2026 by
Excelius-Wang
Contributor
Loading…
fix(metrics): align packed sequence accuracy with ring padding
#10182
opened Sep 17, 2026 by
Excelius-Wang
Contributor
Loading…
fix(sp): exclude internal ring padding from MoE auxiliary loss
#10181
opened Sep 17, 2026 by
Excelius-Wang
Contributor
Loading…
fix(sp): preserve packed boundaries when padding auxiliary fields
#10180
opened Sep 17, 2026 by
Excelius-Wang
Contributor
Loading…
1 task done
fix(template): truncate token type IDs with retained positions
#10176
opened Sep 17, 2026 by
taking-lying-flat
Contributor
Loading…
1 of 4 tasks
fix(gkd): promote low-precision logits before loss computation
#10175
opened Sep 16, 2026 by
taking-lying-flat
Contributor
Loading…
1 of 4 tasks
fix(megatron): honor DPO loss type and normalize IPO logprobs
#10167
opened Sep 16, 2026 by
taking-lying-flat
Contributor
Loading…
1 of 4 tasks
fix(template): reject samples that lose all supervision during truncation
#10166
opened Sep 16, 2026 by
Excelius-Wang
Contributor
Loading…
1 of 4 tasks
fix(gkd): normalize HF loss across accumulation and DDP
#10165
opened Sep 16, 2026 by
taking-lying-flat
Contributor
Loading…
1 of 4 tasks
perf(megatron): remove device syncs in packed CP reconstruction
#10164
opened Sep 16, 2026 by
Ruihan11
Contributor
Loading…
1 of 5 tasks
fix(train): serialize GaLore projector state in checkpoints
#10161
opened Sep 16, 2026 by
MrCapricornLiu
Contributor
Loading…
fix(dataset): invalidate metadata cache on registration
#10160
opened Sep 16, 2026 by
MrCapricornLiu
Contributor
Loading…
fix(train): distinguish activation views during CPU offload
#10157
opened Sep 16, 2026 by
MrCapricornLiu
Contributor
Loading…
1 task done
Previous Next
ProTip!
Filter pull requests by the default branch with base:main.