T Git
首页 开源 帮助 登录 注册
trending/FreeToken
Watch 1
Star 0
Fork 0
mirror of https://github.com/FlashML-org/FreeToken.git synced 2026-11-21 01:52:30 +08:00
Code Issues Packages Projects Releases Wiki Activity
Files
docs/decode-guide
FreeToken/python/freetoken/kernel/csrc/cpu_moe/kernels
T
History
Xiaoze Fan 3d23551c5b perf(cpu-moe): speed up the nvfp4 W4A8 kernels (#851)
The VNNI dots were compute bound per core; with fewer instructions per weight they reach the DRAM read bandwidth.
The W4A8 kernel now follows FREETOKEN_CPU_MOE_ISA.
2026-11-02 15:05:17 -08:00
..
bf16.cpp
refactor(cpu-moe): split the executor into weight formats and gpu sync backends (#670)
2026-10-13 20:12:58 -07:00
ds_fp4.cpp
refactor(cpu-moe): split the executor into weight formats and gpu sync backends (#670)
2026-10-13 20:12:58 -07:00
fp8_block.cpp
refactor(cpu-moe): split the executor into weight formats and gpu sync backends (#670)
2026-10-13 20:12:58 -07:00
gguf.cpp
feat(gguf): apply the weight scales of nvfp4 gguf tensors (#732)
2026-10-17 13:46:50 -07:00
mxfp4.cpp
refactor(cpu-moe): split the executor into weight formats and gpu sync backends (#670)
2026-10-13 20:12:58 -07:00
nvfp4.cpp
perf(cpu-moe): speed up the nvfp4 W4A8 kernels (#851)
2026-11-02 15:05:17 -08:00
📖 帮助文档 · 📡 API · 🔍 开源资源 · 🌐 koklan.biz
T Git · AI 时代的代码协作平台 · © 海南科凯联科技有限公司 · 琼ICP备2025050379号-3