Skip to content

Pull requests: modelscope/ms-swift

New pull request New
95 Open 3,910 Closed
95 Open 3,910 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

feat(megatron): add opt-in chunked full-vocabulary GKD loss
#10114 opened Sep 11, 2026 by taking-lying-flat Contributor Loading…
1 of 4 tasks
fix(megatron): correct rollout weight reload, resume offsets and sampler seed
#10113 opened Sep 11, 2026 by taking-lying-flat Contributor Loading…
1 of 4 tasks
1
surpport deepseek-v4-flash-vision-exp megatron-swift
#10112 opened Sep 11, 2026 by z0o0ey Collaborator Loading…
1 of 4 tasks
Fix Arrow-aligned tool schemas leaking foreign parameters
#10110 opened Sep 11, 2026 by aprylewu Loading…
1 of 4 tasks
1
[Bugfix][RLHF] fix colocate offload and vLLM wake-up ordering
#10109 opened Sep 11, 2026 by rxlqn Loading…
fix(rlhf): align rollout response prefixes during training
#10107 opened Sep 10, 2026 by taking-lying-flat Contributor Loading…
1 of 4 tasks
1
[bugfix][megatron] align GPU/NPU GKD precision
#10102 opened Sep 10, 2026 by xinyu-0877 • Draft
fix(megatron): normalize GKD loss by global valid-token count
#10096 opened Sep 10, 2026 by wzkk123 • Draft
6 of 10 tasks
fix(megatron): gather embedding evaluation metrics across DP ranks
#10091 opened Sep 9, 2026 by MrCapricornLiu Contributor Loading…
1 of 4 tasks
fix(loss): preserve gradients for skipped listwise reranker batches
#10090 opened Sep 9, 2026 by Excelius-Wang Contributor Loading…
fix(metrics): handle period-only text in ROUGE evaluation
#10089 opened Sep 9, 2026 by Excelius-Wang Contributor Loading…
fix(megatron): reduce reranker evaluation metrics across DP ranks
#10088 opened Sep 9, 2026 by Excelius-Wang Contributor Loading…
[NPU] Migrate Megatron training to Core 0.18 with MegatronAdaptor and TENPU
#10087 opened Sep 9, 2026 by addsubmuldiv Collaborator • Draft
2 of 4 tasks
fix(gkd): preserve student gradients for empty token partitions
#10085 opened Sep 9, 2026 by Excelius-Wang Contributor Loading…
fix(trainer): normalize partial gradient accumulation windows
#10084 opened Sep 9, 2026 by MrCapricornLiu Contributor Loading…
1 of 4 tasks
fix(dataset): detect failed streaming packing workers
#10083 opened Sep 9, 2026 by MrCapricornLiu Contributor Loading…
1 of 4 tasks
perf(grpo): batch rollout diagnostic metric gathering
#10082 opened Sep 9, 2026 by taking-lying-flat Contributor Loading…
1 of 4 tasks
fix(loss): compensate InfoNCE gradients for DDP averaging
#10081 opened Sep 9, 2026 by taking-lying-flat Contributor Loading…
1 of 4 tasks
fix(infer): share a background event loop across synchronous requests
#10080 opened Sep 9, 2026 by taking-lying-flat Contributor Loading…
1 of 4 tasks
fix(model): rewrite seq_cls architectures to *ForSequenceClassification
#10063 opened Sep 8, 2026 by RerankerGuo Contributor Loading…
1
feat(rollout): add built-in AgentArk environment support
#10051 opened Sep 7, 2026 by P90-RushB Contributor Loading…
2 of 4 tasks
fix(metrics): mask ignored labels in padding-free seq_acc
#10049 opened Sep 6, 2026 by MaxFreedomPollard Loading…
1 of 4 tasks
docs: add Japanese documents
#10047 opened Sep 5, 2026 by eltociear Contributor Loading…
1 of 4 tasks
Support mixed text and raw-token rollout content in Swift template
#10029 opened Sep 2, 2026 by kerbeans Loading…
2
feat(gkd): add agentic monitoring metrics
#10026 opened Sep 2, 2026 by Casuallkk Loading…
1 of 4 tasks
ProTip! Type g p on any issue or pull request to go back to the pull request listing page.