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/gpt-oss-gateway-startup-timeout 6d02746f36 · Extend GPT-OSS gateway startup timeout · 最后更新于 2026-08-10 07:19:12 +00:00 frostbyte_neo	379 1	ZIP TAR.GZ
feat/k3-dspark-torchspec-draft 5f1e48b5d1 · perf(kimi-k3): fuse per-tap fc_norm · 最后更新于 2026-08-10 04:47:28 +00:00 frostbyte_neo	385 2	ZIP TAR.GZ
towillwu/lora-kernels c1c3cd837a · feat(kernel): Triton LoRA shrink/expand kernels · 最后更新于 2026-08-10 02:37:55 +00:00 frostbyte_neo	383 6	ZIP TAR.GZ
towillwu/lora-moe 65594ed0c4 · feat(moe): integrate MoE LoRA into the Triton bf16 MoE kernel · 最后更新于 2026-08-10 02:37:55 +00:00 frostbyte_neo	383 8	ZIP TAR.GZ
towillwu/lora-serving 5935d2cc5d · feat(lora): LoRA adapter serving runtime · 最后更新于 2026-08-10 02:37:55 +00:00 frostbyte_neo	383 7	ZIP TAR.GZ
towillwu/lora-scheduler 352b19b415 · docs: document the --max-loras scheduler option · 最后更新于 2026-08-10 02:37:40 +00:00 frostbyte_neo	383 5	ZIP TAR.GZ
bot/kimi-k3-aime26-min-score bba0c4ae40 · ci: recalibrate Kimi K3 AIME26 score gate · 最后更新于 2026-08-10 01:10:50 +00:00 frostbyte_neo	383 1	ZIP TAR.GZ
cleanup/mxfp4-drop-dead-decode-paths 99331d4a41 · cleanup(amd): drop dead gfx950 mxfp4 decode entry point and warp-GEMV path · 最后更新于 2026-08-10 01:02:25 +00:00 frostbyte_neo	384 1	ZIP TAR.GZ
perf/gfx950-moe-disable-stage2-atomic-combine 369b9c6c85 · Drop benchmark results in the comment · 最后更新于 2026-08-09 23:52:44 +00:00 frostbyte_neo	384 3	ZIP TAR.GZ
bot/prep-tape-ordered-replay a3cebe9318 · fix(kernel): preserve prep tape operation order · 最后更新于 2026-08-09 23:08:47 +00:00 frostbyte_neo	383 1	ZIP TAR.GZ
lei/refactor-moe a74ab3c459 · refactor(amd): split gfx950 mxfp4 fused.py into fused/ subpackage · 最后更新于 2026-08-09 23:01:52 +00:00 frostbyte_neo	384 1	ZIP TAR.GZ
fix/k3-dspark-mla-cache-dtype d1a225ca48 · fix(kimi-k3): a DSpark draft's MLA cache cannot diverge from the target's · 最后更新于 2026-08-09 11:40:05 +00:00 frostbyte_neo	385 1	ZIP TAR.GZ
fix/k3-attnres-aux-join 5dc5fc3f4a · fix(k3): order the fused AR after the aux-stream attnres partial · 最后更新于 2026-08-09 00:21:24 +00:00 frostbyte_neo	395 1	ZIP TAR.GZ
Yu-Zhewen/gfx1250-situ-ep 21fd97267f · feat(moe): enable gfx1250 SiTU EP dispatch · 最后更新于 2026-08-06 13:32:01 +00:00 frostbyte_neo	421 1	ZIP TAR.GZ
borontion/kernel-cleanup 3c74052159 · move non-standard kernels · 最后更新于 2026-08-05 19:05:37 +00:00 frostbyte_neo	464 9	ZIP TAR.GZ
copilot/fix-typing-errors aea7a8a10b · platform: reject unsupported AMD architectures · 最后更新于 2026-08-01 21:36:02 +00:00 frostbyte_neo	489 4	ZIP TAR.GZ
codex/top-k-logprobs 8c4f35d4dc · Merge branch 'main' into codex/top-k-logprobs · 最后更新于 2026-07-31 22:44:49 +00:00 frostbyte_neo	497 5	ZIP TAR.GZ
perf/kda-cutedsl-fused-decode 7fb90ed1c6 · perf(kda): CuTe DSL fused decode for KDA linear attention · 最后更新于 2026-07-30 20:41:54 +00:00 frostbyte_neo	520 1	ZIP TAR.GZ
flamingo/m3-fused-cache-insert 66d514f0cb · perf: fuse MiniMax sparse cache insertion · 最后更新于 2026-07-29 11:29:48 +00:00 frostbyte_neo	524 1	ZIP TAR.GZ
zt/dev/mm 2a3f57e9a1 · Merge branch 'main' into zt/dev/mm · 最后更新于 2026-07-29 06:21:36 +00:00 frostbyte_neo	526 2	ZIP TAR.GZ

... 9 10 11 12 13 ...

默认分支

分支列表