# MemVanta trained-model layer profile
prefill_ms=90914.9 decode_total_ms=8813.2 sampling_ms=0.365736
prefill_cycles_per_token=1736919714.74 decode_cycles_per_token=1347001329.62
qkv_projection_ms=23687.70 attention_output_projection_ms=7905.51 ffn_gemm_ms=63707.33 output_head_ms=3202.24
prefill_qkv_projection_ms=22297.35 prefill_attention_output_projection_ms=7431.91 prefill_ffn_gemm_ms=59938.14 prefill_output_head_ms=375.35 prefill_kernel_ms=90042.75
decode_qkv_projection_ms=1390.35 decode_attention_output_projection_ms=473.60 decode_ffn_gemm_ms=3769.19 decode_output_head_ms=2826.89 decode_kernel_ms=8460.03
non_gemm_core_ms=1224.96 (attention softmax/KV + RMSNorm + RoPE + elementwise/residual)
layer,qkv_ms,o_proj_ms,ffn_ms,total_projection_ms,pct_kernel_ms,projection_ms_per_token
0,742.05,247.02,1989.84,2978.90,3.02,20.69
1,739.53,247.29,1991.62,2978.44,3.02,20.68
2,739.65,246.65,1989.08,2975.38,3.02,20.66
3,739.44,246.93,1990.46,2976.83,3.02,20.67
4,741.71,247.29,1991.00,2980.00,3.03,20.69
5,739.27,246.73,1990.11,2976.11,3.02,20.67
6,739.27,246.78,1988.83,2974.88,3.02,20.66
7,739.18,246.77,1989.59,2975.54,3.02,20.66
8,739.55,246.65,1989.25,2975.45,3.02,20.66
9,738.94,246.90,1990.29,2976.13,3.02,20.67
10,739.09,246.94,1988.82,2974.85,3.02,20.66
11,743.59,246.74,1989.87,2980.19,3.03,20.70
12,739.27,246.61,1991.34,2977.22,3.02,20.68
13,739.79,247.12,1991.83,2978.74,3.02,20.69
14,739.47,246.68,1991.05,2977.20,3.02,20.67
15,740.26,247.00,1989.42,2976.68,3.02,20.67
16,739.37,246.91,1989.22,2975.51,3.02,20.66
17,739.13,246.81,1990.16,2976.11,3.02,20.67
18,739.92,247.26,1990.88,2978.07,3.02,20.68
19,740.45,246.99,1991.71,2979.15,3.02,20.69
20,740.87,247.42,1991.58,2979.88,3.03,20.69
21,740.34,247.12,1991.91,2979.36,3.02,20.69
22,739.91,247.31,1993.15,2980.37,3.03,20.70
23,740.26,248.41,1994.13,2982.80,3.03,20.71
24,739.88,247.07,1991.17,2978.11,3.02,20.68
25,742.03,247.27,1990.90,2980.21,3.03,20.70
26,740.48,247.80,1992.30,2980.58,3.03,20.70
27,743.84,247.04,1993.47,2984.35,3.03,20.72
28,739.93,246.88,1991.74,2978.55,3.02,20.68
29,740.51,247.08,1992.32,2979.91,3.03,20.69
30,740.75,246.90,1990.27,2977.91,3.02,20.68
31,739.97,247.12,1990.04,2977.13,3.02,20.67
profile_csv=sevenb-ffn-tuning/q8act-b64.csv
