Skip to content

Pull requests: lightseekorg/tokenspeed

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
Assignee
Filter by who’s assigned
Assigned to nobody Loading
Sort

Pull requests list

perf(kimi3): let the MoE tail join its reductions without a lane
#1330 opened Aug 31, 2026 by nirvedhmeshram Contributor Loading…
perf(k3): optimize MXFP4 MoE prefill on gfx950
#1328 opened Aug 31, 2026 by qedawkins Contributor Loading…
test: update k3 agentic bench
#1326 opened Aug 31, 2026 by syuoni Member Draft
perf(k3): optimize latent MoE tail scheduling and finalize
#1324 opened Aug 31, 2026 by chenht2022 Contributor Loading…
fix(moe): pad NVFP4 expert shards for TRT-LLM
#1322 opened Aug 31, 2026 by tuanzhangCS Contributor Loading…
perf(k3): decode gates, GEMV route QA, and FlashInfer 0.6.18 re-tune
#1318 opened Aug 31, 2026 by nperrin-fr Collaborator Loading…
ci: fix B300 DeepSWE runtime access
#1317 opened Aug 30, 2026 by lightseek-bot Contributor Draft
perf(kda): use FlashInfer fused decode on Blackwell
#1311 opened Aug 30, 2026 by chenht2022 Contributor Draft
feat(glm-5.3-flash): add day-0 serving support high priority
#1259 opened Aug 26, 2026 by FlamingoPg Contributor Loading…
[WIP] feat: add DFlash2 draft serving support
#1211 opened Aug 23, 2026 by torchspec-bot Collaborator Draft
feat(dspark): expose forced-rejection diagnostics
#1170 opened Aug 20, 2026 by jhinpan Contributor Loading…
fix(cache): route null decode pages to dummy slot
#1169 opened Aug 20, 2026 by jhinpan Contributor Loading…
feat(mla): add decode context parallelism
#1162 opened Aug 20, 2026 by qimcis Contributor Loading…
ProTip! Updated in the last three days: updated:>2026-08-28.