Skip to content

Pull requests: sgl-project/SpecForge

New pull request New
53 Open 361 Closed
53 Open 361 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

Expose flex_attention kernel options in DFlash and Domino training
#586 opened Jun 18, 2026 by heiheiha798 Loading…
11
[Feature] VLM DFlash Training: Multi-Model Support for Qwen3-VL / Qwen3.5 / Qwen3.6
#585 opened Jun 18, 2026 by zyk42 Loading…
1
Add Ascend NPU support for the Domino training pipeline (Qwen3.5-4B)
#584 opened Jun 18, 2026 by curnane-lab Loading…
1
feat: Add Ascend npu support.
#576 opened Jun 9, 2026 by menogrey Loading…
6 tasks
1
8
Add P-EAGLE training support
#575 opened Jun 4, 2026 by thyways Contributor Loading…
3 of 6 tasks
1
1
feat: Train-Inference Disaggregation for Remote Target Model Serving
#573 opened Jun 2, 2026 by moehanabi Contributor Loading…
4 of 6 tasks
12
EAGLE3.1 Support
#568 opened May 27, 2026 by bluecoffee8 Loading…
6 tasks
14
feat: add automatic device detection for non-CUDA backends
#559 opened May 25, 2026 by curnane-lab Loading…
6 tasks done
6
Support sharded target logits for EAGLE3 online training
#558 opened May 25, 2026 by Yukino256 Loading…
6 tasks
9
Supports eagle3 training for Gemma3 27B and Gemma4 26B.
#553 opened May 1, 2026 by pyc96 Collaborator Loading…
6 tasks
4
fix benchmark template
#552 opened Apr 30, 2026 by Dogacel Loading…
6 tasks
1
1
Add transformers-like checkpoint parameters (--save-total-limit, --save-strategy, and so on)
#547 opened Apr 27, 2026 by thechaos16 Loading…
2 of 6 tasks
7
feat: add is_vlm param to safe_conversations_generator for multimodal data support
#545 opened Apr 24, 2026 by sunny-infra Loading…
2 of 6 tasks
5
chore: regenerate_train_data accepts API key and https URL
#543 opened Apr 24, 2026 by lianakoleva Loading…
1 of 6 tasks
1
add the configs for qwen3-vl-8b-instruct model
#542 opened Apr 23, 2026 by sunny-infra Loading…
1 of 6 tasks
4
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: 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
ProTip! Add no:assignee to see everything that’s not assigned.