lightseekorg--tokenspeed

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

perf/v4-trtllm-mhc-decode-gating e7eceb73b3 · fix(deepseek-v4): gate experimental trtllm mhc backend · 最后更新于 2026-07-07 12:40:15 +00:00 frostbyte_neo	738 2	ZIP TAR.GZ
ci/deepseek-v4-mtp-aime25 4727759f22 · ci: stabilize DeepSeek V4 MTP AIME25 eval · 最后更新于 2026-07-06 01:18:39 +00:00 frostbyte_neo	736 9	ZIP TAR.GZ
perf/v4-trtllm-mhc dc2fbb50c2 · perf(deepseek-v4): integrate TRT-LLM mHC kernels · 最后更新于 2026-06-28 13:31:19 +00:00 frostbyte_neo	738 1	ZIP TAR.GZ

分支列表