Skip to content

Pull requests: lightseekorg/tokenspeed

New pull request New
42 Open 707 Closed
42 Open 707 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(multi-node): harden Qwen3.5 multi-node execution
#780 opened Jul 23, 2026 by tuanzhangCS Contributor Loading…
3
fix(qwen3.5): allow MTP to be quantized
#777 opened Jul 23, 2026 by minedec Contributor Loading…
2
fix: Support NVIDIA Nemotron-3-Super-120B-A12B
#770 opened Jul 22, 2026 by lian125537 Loading…
2
feat(kernel): Add validation per family/mode for kernel registration
#769 opened Jul 22, 2026 by Max191 Contributor Loading…
6
Support MiniMax M3 CPU KVStore
#758 opened Jul 22, 2026 by FlamingoPg Contributor • Draft
[WIP] feat(deepseek): land V4 L1 KV cache on heterogeneous flat kv pools
#750 opened Jul 21, 2026 by SimonCqk Contributor Loading…
4
[AMD] Add initial support for pipelined GFX1250 Gluon MHA decode
#746 opened Jul 21, 2026 by adityakankariya Loading…
10
refactor(runtime): make ForwardContext tensor-free (restore "Do not contain Tensor")'
#745 opened Jul 21, 2026 by rjzhb Collaborator • Draft
2 of 5 tasks
feat(profile): Link VizTracer to Proton CPU scopes
#740 opened Jul 20, 2026 by antiagainst Member • Draft
1
feat(lora): LoRA adapter serving
#738 opened Jul 20, 2026 by qywu Collaborator Loading…
feat(moe): R3 rollout routing replay — slot-indexed expert-routing pool (scaffold)
#737 opened Jul 20, 2026 by qywu Collaborator • Draft
1
feat(engine): return_token_ids — echo raw prompt/output token ids (RL anti-drift)
#736 opened Jul 20, 2026 by qywu Collaborator • Draft
feat(scheduler): per-adapter KV prefix-cache namespace + max_loras batch cap
#735 opened Jul 19, 2026 by qywu Collaborator Loading…
[wip] refactor(kernel): migrate GDN Triton kernels to tensor descriptors
#721 opened Jul 18, 2026 by raikonenfnu Contributor Loading…
4 tasks done
2
[WIP][AMD] Implement MTP support for qwen3.5 MXFP4
#720 opened Jul 18, 2026 by raikonenfnu Contributor Loading…
3 tasks
3
perf(deepseek-v4): capture DSA projections and cache writes into the prefill graph
#714 opened Jul 17, 2026 by dongjiyingdjy Contributor Loading…
feat(scheduler): live-tail allocation for sliding groups
#694 opened Jul 16, 2026 by nperrin-fr Collaborator • Draft
4
fix(scheduler): check host write-back capacity before mutating retrac…
#688 opened Jul 15, 2026 by alexps9 Loading…
3 tasks done
2
feat(rl): stamp weight versions on generation responses
#672 opened Jul 14, 2026 by HJSang Collaborator Loading…
5 tasks done
3
perf(deepseek-v4): select two-stage fused mHC launch by token count
#661 opened Jul 13, 2026 by Xiangyi1996 Collaborator • Draft
fix(pd): correct DeepSeek V4 layerwise cache handoff
#660 opened Jul 13, 2026 by lucifer1004 Contributor • Draft
feat(runtime): scheduler-driven full_refresh bit for page-table mirror
#651 opened Jul 12, 2026 by raikonenfnu Contributor Loading…
13
feat(deepseek-v4): enable SM120 serving
#648 opened Jul 11, 2026 by lucifer1004 Contributor • Draft
feat(kernel): add SM120 FlashInfer MXFP4 MoE
#645 opened Jul 11, 2026 by lucifer1004 Contributor Loading…
11
fix(deepseek-v4): route large-token fused mHC around the allinone cliff
#622 opened Jul 9, 2026 by Xiangyi1996 Collaborator • Draft
2
ProTip! What’s not been updated in a month: updated:<2026-06-23.