github-featured / lightseekorg--tokenspeed

分支列表

main

0ba6b6aab7 · perf(mla): add opt-in packed Q decode for FP8 and FP16 (#1512) · 最后更新于 2026-09-12 09:04:01 +00:00

bot/glm-5-3-flash-gpu-memory-utilization-admission-budget 2d497754ff · ci(glm53): lower memory ratio for scheduler stress · 最后更新于 2026-09-08 00:56:55 +00:00 frostbyte_neo	83 3	ZIP TAR.GZ
bot/scheduler-retraction-liveness 3f5a77fcbd · refactor(scheduler): judge the retraction exemption by the admission-time budget · 最后更新于 2026-09-07 18:04:31 +00:00 frostbyte_neo	83 2	ZIP TAR.GZ
bot/glm-5-3-flash-gpu-memory-utilization-control 56ef6eb1ff · ci(glm53): lower memory ratio for scheduler stress · 最后更新于 2026-09-07 17:31:40 +00:00 frostbyte_neo	83 1	ZIP TAR.GZ
bot/glm-5-3-flash-gpu-memory-utilization-liveness-fix fe78df0b8d · ci(glm53): lower memory ratio for scheduler stress · 最后更新于 2026-09-07 17:31:40 +00:00 frostbyte_neo	83 2	ZIP TAR.GZ
refactor/state-family-is-checkpoint-shape 4fdbbae6f3 · refactor(scheduler): a State cache group is the checkpoint shape · 最后更新于 2026-09-07 17:18:13 +00:00 frostbyte_neo	83 4	ZIP TAR.GZ
hongtaoc/k3-flashinfer-exact-autotune cb7597c7c9 · fix(autotune): preserve eager phase during decode tuning · 最后更新于 2026-09-07 17:14:54 +00:00 frostbyte_neo	82 6	ZIP TAR.GZ
zt/dev/sparse_attn_opt_bf16 770a204a23 · feat(qsa): support TP1 and TP2 CuTe head groups · 最后更新于 2026-09-07 12:51:25 +00:00 frostbyte_neo	83 3	ZIP TAR.GZ
k3-latent-down-column-shard 92afe1de8a · Merge branch 'main' into k3-latent-down-column-shard · 最后更新于 2026-09-07 04:59:38 +00:00 frostbyte_neo	83 30	ZIP TAR.GZ
lei/update-triton-0906 9b457ac036 · deps: update Triton/Proton to post20260906 · 最后更新于 2026-09-06 21:35:40 +00:00 frostbyte_neo	89 1	ZIP TAR.GZ
jjd/fix_mixed_batch_draft b0bdf623dc · fix: preserve TRTLLM draft prefill metadata during decode refresh · 最后更新于 2026-09-06 19:27:53 +00:00 frostbyte_neo	89 1	ZIP TAR.GZ
lei/try-triton cb61c23ed0 · Update TokenSpeed Triton vendor release · 最后更新于 2026-09-06 17:20:53 +00:00 frostbyte_neo	89 9	ZIP TAR.GZ
unify-state-growth-reserve afdd2c2a1f · Merge branch 'main' into unify-state-growth-reserve · 最后更新于 2026-09-06 04:22:50 +00:00 frostbyte_neo	89 3	ZIP TAR.GZ
borontion/fix-mi450-sim-runtime 58b28b20ad · Merge origin/main into borontion/fix-mi450-sim-runtime · 最后更新于 2026-09-05 23:43:24 +00:00 frostbyte_neo	91 5	ZIP TAR.GZ
borontion/diagnose-kimi-k3-evalscope-1.9.1 1a082199e7 · test(ci): pin EvalScope for Kimi K3 diagnosis · 最后更新于 2026-09-05 16:46:03 +00:00 frostbyte_neo	104 1	ZIP TAR.GZ
jjd/amd-k3-eagle3-g16-fix-ed6f83dc 7a41647396 · ci(amd): isolate ed6 Kimi-K3 comparison · 最后更新于 2026-09-04 10:10:12 +00:00 frostbyte_neo	100 3	ZIP TAR.GZ
zt/qsa_sparse_cutedsl 9a89f877b0 · fix(qsa): require metadata capacity explicitly · 最后更新于 2026-09-04 10:09:54 +00:00 frostbyte_neo	99 10	ZIP TAR.GZ
jjd/amd-k3-eagle3-g16-fix-final-2903 c0f0b49f17 · test(amd): isolate gdot reduction output buffer · 最后更新于 2026-09-04 10:04:29 +00:00 frostbyte_neo	104 3	ZIP TAR.GZ
jjd/amd-k3-eagle3-fix-gdot-reduce-output-ci 237a5928e5 · ci(amd): isolate Kimi-K3 reduction fix validation · 最后更新于 2026-09-04 09:05:02 +00:00 frostbyte_neo	99 2	ZIP TAR.GZ
jjd/amd-k3-eagle3-gdot-g16-private-reduce a507b430dd · test(amd): isolate gdot reduction output buffer · 最后更新于 2026-09-04 08:58:36 +00:00 frostbyte_neo	107 4	ZIP TAR.GZ
jjd/amd-k3-eagle3-fix-gdot-output-ci 3cef620d26 · ci(amd): isolate Kimi-K3 Eagle3 fix validation · 最后更新于 2026-09-04 08:39:20 +00:00 frostbyte_neo	99 2	ZIP TAR.GZ

1 2 3 4 5 ...

默认分支

分支列表