auto-round
ff76f390
- opt 16 bit group gemm
Go
Login via GitHub
Home
Pricing
FAQ
Install
Login
via GitHub
Commit
View On
GitHub
Commit
23 days ago
opt 16 bit group gemm Signed-off-by: Dong, Bo1 <bo1.dong@intel.com>
References
opt_moe_kernel
#2090 - Opt the moe performance
#2108 - ark.moe: switch auto dispatch to total-token threshold + env override
#2111 - Optimize INT4 MoE prefill/decode (sym) XPU kernels with dedicated w4a16 tiling
#2143 - feat: W4A8 ARK XPU MoE kernel (int4 weight / int8 compute) with prefill + decode
Author
a32543254
Committer
a32543254
Parents
8be551d3
Loading