This commit is contained in:
yadaish
2025-11-28 11:31:25 +00:00
parent 0642090655
commit 4aa2407cf0

View File

@@ -36,7 +36,7 @@ float invoke_a16w4_moe_gemm(int n_warmup, int n_repeat, const MoeHostArgs& args)
std::size_t flop = std::size_t(2) * args.M * args.N * args.K;
std::size_t num_byte = sizeof(ADataType) * args.M * args.K +
sizeof(BDataType) * args.N * args.K / PackedSize +
sizeof(BDataType) * args.N * args.K * std::min(args.experts, args.NumTokens * args.TopK) / PackedSize +
sizeof(CDataType) * args.M * args.N;
float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
float gb_per_sec = num_byte / 1.E6 / ave_time;