-
Notifications
You must be signed in to change notification settings - Fork 321
Pull requests: sgl-project/SpecForge
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
feat: HYBRID_SHARD draft training across colocated target islands
#784
opened Aug 25, 2026 by
maocheng23
Collaborator
•
Draft
3 of 4 tasks
feat: colocated online training with in-process SGLang capture
#783
opened Aug 25, 2026 by
maocheng23
Collaborator
•
Draft
4 tasks done
fix: create the metrics tracker only on trainer rank zero
#782
opened Aug 25, 2026 by
maocheng23
Collaborator
•
Draft
3 tasks done
fix: abort process groups on rank-local training failure
#781
opened Aug 25, 2026 by
maocheng23
Collaborator
•
Draft
3 tasks done
refactor: extract packed capture_rows from the offline SGLang backend
#780
opened Aug 25, 2026 by
maocheng23
Collaborator
•
Draft
3 tasks done
perf(rocm): enable Liger fused RMSNorm/SwiGLU for AMD DFlash configs
#779
opened Aug 20, 2026 by
zhentaocc
Loading…
3 tasks done
fix(data): skip zero-message conversations before apply_chat_template
#777
opened Aug 20, 2026 by
zhentaocc
Loading…
fix(patch): repair spec-capture-ascend-mount.patch hunk headers
#776
opened Aug 20, 2026 by
curnane-lab
Contributor
Loading…
2 of 6 tasks
fix(dflash): gate flex_attention BACKEND=TRITON on torch>=2.11 (fixes ROCm)
#773
opened Aug 19, 2026 by
zhentaocc
Loading…
feat: --trim-backbone-rows — run TTT steps >= 1 only on rows that can still emit loss
#771
opened Aug 19, 2026 by
julyanghar
Contributor
•
Draft
4 tasks done
feat: Intel XPU (Arc Pro B70) training support for DSpark drafters
#769
opened Aug 16, 2026 by
rmacy
Loading…
Add configurable GQA and MLA sliding-window draft layers
#767
opened Aug 15, 2026 by
maocheng23
Collaborator
•
Draft
feat: add scalable colocated online training
#766
opened Aug 12, 2026 by
maocheng23
Collaborator
•
Draft
4 of 5 tasks
feat(scripts): build a draft vocabulary mapping without regenerating features
#765
opened Aug 12, 2026 by
fg11991
Loading…
feat(dspark): support a pruned draft vocabulary for the DFlash family
#764
opened Aug 12, 2026 by
fg11991
Loading…
Add online-live mode: train drafters from production serving traffic
#763
opened Aug 10, 2026 by
sherlockwu
Collaborator
Loading…
Add config-selectable MLA draft attention
#758
opened Aug 8, 2026 by
maocheng23
Collaborator
Loading…
fix: degrade empty anchor sampling to a masked batch to avoid multi-r…
#737
opened Aug 1, 2026 by
Boreas618
Contributor
Loading…
2 of 6 tasks
[Feature] Multimodal (image+text) DFlash training via server-side capture (Qwen3.5)
#730
opened Jul 29, 2026 by
curnane-lab
Contributor
Loading…
3 of 6 tasks
feat(dspark): add Ulysses sequence parallel training
#728
opened Jul 28, 2026 by
maocheng23
Collaborator
•
Draft
4 of 5 tasks
feat(dspark): add Qwen3.5-4B draft config and managed-local NPU disaggregated recipe
#727
opened Jul 28, 2026 by
curnane-lab
Contributor
•
Draft
4 of 6 tasks
feat(domino): add Qwen3.5-4B managed-local NPU disaggregated recipe
#726
opened Jul 28, 2026 by
curnane-lab
Contributor
Loading…
1 of 6 tasks
feat(dspark): add offline USP training support
#725
opened Jul 28, 2026 by
calyoung80
Loading…
2 of 6 tasks
Previous Next
ProTip!
Updated in the last three days: updated:>2026-08-22.