Skip to content

Pull requests: lightseekorg/tokenspeed

New pull request New
15 Open 151 Closed
15 Open 151 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(deepseek-v4): corrected profiling to estimate cache capacity.
#173 opened May 17, 2026 by SimonCqk Contributor Loading…
@SimonCqk
perf(moe): triton biased grouped topk for deepseek-v3 routing
#171 opened May 17, 2026 by roycho96 Contributor Loading…
1
wip: EAGLE post-norm
#170 opened May 17, 2026 by Dogacel • Draft
fix: detect GPU arch automatically for kernel building
#169 opened May 17, 2026 by Dogacel Loading…
1
[WIP] perf: add gluon fp16 prefill kernel
#165 opened May 16, 2026 by borontion Contributor Loading…
13
feat(kvstore): support mamba l2 cache transfers
#162 opened May 15, 2026 by XucSh Contributor Loading…
Perf[Qwen3.5]: eliminate Mamba intermediate state memcpy in MTP target-verify
#159 opened May 15, 2026 by tuanzhangCS Contributor • Draft
[WIP] feat(deepseek-v4): support prefix cache snapshots
#146 opened May 14, 2026 by SimonCqk Contributor Loading…
perf(sampling): opt-in fast verify path for topk=1 chain spec
#133 opened May 13, 2026 by cicirori • Draft
4 tasks done
[Draft]feat(deepseek-v4): support MTP speculative decoding
#123 opened May 13, 2026 by dongjiyingdjy Contributor Loading…
[WIP] feat(lora): LoRA adapter serving
#83 opened May 11, 2026 by qywu Collaborator • Draft
1 of 7 tasks
perf(qwen3): cut H100 decode kernel time -8% with fused stride-aware kernels high priority
#81 opened May 11, 2026 by qywu Collaborator Loading…
5 of 7 tasks
@borontion
2
fix: retraction load back race condition.
#74 opened May 11, 2026 by LorrinWWW Contributor Loading…
perf: chunked-prefill prefix cache update for non-hybrid models
#22 opened May 7, 2026 by LorrinWWW Contributor Loading…
@LorrinWWW
1
fix: wait per-layer on drafter KV pool during cpu cache loadback
#6 opened May 6, 2026 by LorrinWWW Contributor Loading…
@LorrinWWW
ProTip! Follow long discussions with comments:>50.