Skip to content

Pull requests: deepseek-ai/DeepSpec

New pull request New
26 Open 12 Closed
26 Open 12 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

fix: narrow bare except to ImportError in constant module init
#69 opened Jul 17, 2026 by fix2015 Loading…
Fix: Feat: Compatible with Gamme4 MoE structure
#68 opened Jul 11, 2026 by MayVerse4 Loading…
1
Publish dataset splits atomically
#67 opened Jul 11, 2026 by morluto Loading…
Avoid duplicate final checkpoint save
#66 opened Jul 11, 2026 by morluto Loading…
Add Sequential Temperature Scaling for confidence-head calibration
#63 opened Jul 9, 2026 by DaoyuanLi2816 Contributor Loading…
[DFlash/DSpark] Long-context training via RoPE scaling
#59 opened Jul 8, 2026 by zhangwenhe1007 Loading…
[DFlash/DSpark] Optional direct-acceptance (LK) loss
#58 opened Jul 8, 2026 by zhangwenhe1007 Loading…
Reduce chat data exposure in generation and dataset handling
#55 opened Jul 6, 2026 by MayVerse4 Loading…
build: add pyproject.toml for pip/uv installable packaging
#54 opened Jul 6, 2026 by zhangxin81 Loading…
Add Ministral3 DeepSpec reproduction experiments
#50 opened Jul 4, 2026 by andylolu2 • Draft
Add distribution-exactness tests for speculative rejection sampling
#48 opened Jul 2, 2026 by whycoming Loading…
feat: add chunked gather for anchor hidden state alignment
#47 opened Jul 2, 2026 by PHOEBEMOON0802 Loading…
Add target cache size estimator and smoke-test docs
#42 opened Jul 1, 2026 by p9966 Loading…
Add an sdpa attention path for DSpark training on consumer GPUs
#41 opened Jul 1, 2026 by DaoyuanLi2816 Contributor Loading…
Fix DataLoader crash when data.num_workers=0
#40 opened Jul 1, 2026 by DaoyuanLi2816 Contributor Loading…
Respect CUDA_VISIBLE_DEVICES in data scripts
#38 opened Jul 1, 2026 by Oxygen56 Loading…
1
Fix speculative sampling, scheduler, mask, and loss metric correctness
#30 opened Jun 30, 2026 by morluto Loading…
Prevent training data corruption during generation resumptions
#29 opened Jun 30, 2026 by morluto Loading…
feat(data): add support for FP8 target cache quantization (float8_e4m3fn & float8_e5m2)
#27 opened Jun 30, 2026 by ApurveKaranwal Loading…
1
1
test
#25 opened Jun 30, 2026 by ManuReddy17 Loading…
Add online training mode for DSpark (no precomputed target cache)
#23 opened Jun 30, 2026 by Ofir408 Loading…
1
2
Load target embeddings without full model init
#11 opened Jun 29, 2026 by zhangxin81 Loading…
3
test: add pytest suite + CI; fix empty-shard crash in JsonLineDataset
#10 opened Jun 28, 2026 by ajinkyajawale14499 Loading…
1
[feature] Add Ascend NPU Support While Preserving GPU Compatibility
#9 opened Jun 28, 2026 by sunny-infra Loading…
6 tasks done
11
Add pyproject.toml with project metadata, dependencies, and build conifg
#7 opened Jun 28, 2026 by Seas0 Loading…
1
ProTip! Add no:assignee to see everything that’s not assigned.