# test bench_moe_kernel_tiling.py kernel_tiling: python3 bench_moe_kernel_tiling.py \ --hidden_size 7168 \ --intermediate_size 2048 \ --num_experts_per_tok 8 \ --expert_num 256 \ --max_len 51200 \ --layer_num 1 \ --qlen 1024 \ --quant int8 \ --warm_up_iter 500 \ --test_iter 1000 \ --threads 160 \ --m_block 320 \ # --n_block_up_gate 256 \ # --n_block_down 128 \ # --n_block_up_gate_prefi 256 \ # --n_block_down_prefi 128 \ # --n_block_up_gate 256 \ # --n_block_down 512 \