Skip to content

Pull requests: sgl-project/SpecForge

New pull request New
47 Open 335 Closed
47 Open 335 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

support quant
#540 opened Apr 17, 2026 by liusy58 Loading…
6 tasks
2
Fix/gemma3 eagle3 hooks
#538 opened Apr 17, 2026 by tcligg • Draft
6 tasks
1
fix: correct vocab_size to 262144 in gemma3-1b-eagle3 configuration
#537 opened Apr 16, 2026 by javierlimt6 Loading…
1 of 6 tasks
1
Fix incorrect LSE gradients in cached FlashAttention for Eagle training
#536 opened Apr 16, 2026 by uygnef Collaborator Loading…
6 tasks
4
fix: EAGLE-3 training compatibility with multimodal-wrapped targets and large vocabs
#535 opened Apr 16, 2026 by elad-inferize Loading…
4 tasks done
4
Preserve drafter vocab mapping when fine-tuning from a checkpoint
#534 opened Apr 15, 2026 by luv-bansal Loading…
2
[Fix] preserve image data in preprocess for VLM training on multimodal data
#532 opened Apr 13, 2026 by jamesahou Loading…
2 of 6 tasks
1
feat: add EAGLE3 support for Step-3.5-Flash
#530 opened Apr 13, 2026 by zijiexia Loading…
3 tasks
2
fix: Bump sglang version from 0.5.9 to 0.5.10
#529 opened Apr 13, 2026 by moehanabi Contributor Loading…
1 of 6 tasks
5
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
2
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
9
[Feature] Train infer disaggregated
#523 opened Apr 2, 2026 by jiapingW Collaborator Loading…
5 tasks
7
fix: Make template override arg work correctly
#522 opened Apr 1, 2026 by moehanabi Contributor Loading…
1 of 6 tasks
4
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 Collaborator • Draft
6 tasks
4
DFlash VLM training support with SGLang backend
#505 opened Mar 17, 2026 by Mandy3311 Contributor 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 Contributor 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
11
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
ProTip! Exclude everything labeled bug with -label:bug.