Skip to content

Pull requests: sgl-project/SpecForge

New pull request New
39 Open 333 Closed
39 Open 333 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

Reduce peak GPU memory in Eagle3 online target generation by avoiding an extra logits copy
#528 opened Apr 9, 2026 by zijiexia Loading…
1 of 6 tasks
1
Fix VLM preprocessing and add mRoPE position handling in target head
#527 opened Apr 8, 2026 by liusy58 Loading…
6 tasks
4
Fix multimodal hidden-state preparation for Qwen3-VL models
#526 opened Apr 8, 2026 by liusy58 Loading…
6 tasks
4
support qwen3vl-235b
#525 opened Apr 8, 2026 by liusy58 Loading…
6 tasks
3
feat: reduce Eagle3 training memory spike via all-to-all sharding
#524 opened Apr 5, 2026 by laoconeth Loading…
2 of 6 tasks
1
8
[Feature] Train infer disaggregated
#523 opened Apr 2, 2026 by jiapingW Loading…
5 tasks
7
fix: Make template override arg work correctly
#522 opened Apr 1, 2026 by moehanabi Loading…
1 of 6 tasks
4
fix: Support different version of PCG args
#517 opened Mar 30, 2026 by moehanabi Loading…
1 of 6 tasks
5
feat: add Qwen3.5 Dense Model EAGLE3 training support
#516 opened Mar 28, 2026 by 36330 Loading…
3
[draft] sp for dflash
#507 opened Mar 19, 2026 by uygnef • Draft
6 tasks
4
DFlash VLM training support with SGLang backend
#505 opened Mar 17, 2026 by Mandy3311 Loading…
22
Fix: support preformatted text datasets in train_eagle3.py (avoid forcing conversations generator)
#498 opened Mar 9, 2026 by Seun-Ajayi Loading…
6 tasks
4
[Feature] Support DFlash Speculative Decoding Training for Qwen3.5 Models
#495 opened Mar 7, 2026 by EanWang211123 Loading…
6 tasks
2
feat: add GLM-4.7-Flash EAGLE3 training support
#493 opened Mar 6, 2026 by lujangus Loading…
3 tasks
8
Add LK Losses: Direct Acceptance Rate Optimization for Speculative Decoding
#492 opened Mar 6, 2026 by MrShevan Loading…
4 tasks done
10
Fix OOM from glibc malloc arena retention during large model training
#491 opened Mar 4, 2026 by yifjiang • Draft
4 tasks
1
Fix eval dataset race condition in multi-node DP training
#490 opened Mar 4, 2026 by yifjiang • Draft
3 tasks
1
Fix multi-node DP training crash from FlashInfer CUDA IPC handles
#489 opened Mar 4, 2026 by yifjiang Loading…
3 tasks done
2
fix: correct always-true condition in HarmonyParser causing duplicate system prompts
#484 opened Mar 2, 2026 by Bias92 Loading…
1
fix: use processor.tokenizer for apply_chat_template in VLM preprocessing
#480 opened Mar 2, 2026 by Bias92 Loading…
2 tasks
1
4
Add Qwen3-VL training adaptation for DFlash
#475 opened Feb 28, 2026 by gq112 • Draft
6 tasks done
4
(feat)Draft model supports Qwen3 MoE
#468 opened Feb 13, 2026 by sadasdasdasdasasd Loading…
1 of 6 tasks
1
add resume for dflash training script
#460 opened Feb 4, 2026 by tianhaoz95 Loading…
1 of 6 tasks
3
feat: add MiniCPM-V-4 VLM support for EAGLE3 training
#451 opened Jan 26, 2026 by Lollipop Loading…
4
dLLM(Dflash) Offline Training Support
#445 opened Jan 21, 2026 by Ximingwang-09 Loading…
6 tasks
4
ProTip! Type g p on any issue or pull request to go back to the pull request listing page.