# MemVanta trained-model layer profile
prefill_ms=44940.7 decode_total_ms=8846.77 sampling_ms=0.364831
prefill_cycles_per_token=858586768.65 decode_cycles_per_token=1352132052.69
qkv_projection_ms=12004.80 attention_output_projection_ms=4077.36 ffn_gemm_ms=33274.91 output_head_ms=3207.71
prefill_qkv_projection_ms=10615.19 prefill_attention_output_projection_ms=3592.48 prefill_ffn_gemm_ms=29484.69 prefill_output_head_ms=375.62 prefill_kernel_ms=44067.98
decode_qkv_projection_ms=1389.61 decode_attention_output_projection_ms=484.87 decode_ffn_gemm_ms=3790.23 decode_output_head_ms=2832.09 decode_kernel_ms=8496.80
non_gemm_core_ms=1222.30 (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
0,375.62,123.85,1043.16,1542.62,2.93,10.71
1,375.30,124.65,1055.87,1555.81,2.96,10.80
2,376.58,126.56,1033.95,1537.08,2.92,10.67
3,381.95,125.55,1056.78,1564.28,2.98,10.86
4,375.62,126.55,1056.65,1558.82,2.97,10.83
5,384.79,129.21,1034.85,1548.85,2.95,10.76
6,378.33,127.13,1054.71,1560.16,2.97,10.83
7,375.58,128.24,1026.51,1530.34,2.91,10.63
8,371.83,126.27,1066.04,1564.15,2.98,10.86
9,378.14,128.18,1040.67,1546.99,2.94,10.74
10,371.13,126.52,1056.47,1554.13,2.96,10.79
11,374.41,127.38,1011.50,1513.29,2.88,10.51
12,374.06,128.14,1048.03,1550.23,2.95,10.77
13,378.74,125.62,1017.46,1521.82,2.90,10.57
14,381.13,126.53,1019.81,1527.47,2.91,10.61
15,366.54,126.83,1021.52,1514.89,2.88,10.52
16,372.33,127.97,1042.18,1542.48,2.93,10.71
17,370.69,126.06,1037.23,1533.99,2.92,10.65
18,371.17,128.81,1025.67,1525.65,2.90,10.59
19,376.83,129.82,1038.56,1545.20,2.94,10.73
20,374.10,127.11,1037.69,1538.90,2.93,10.69
21,377.56,128.27,1058.41,1564.24,2.98,10.86
22,376.30,128.83,1028.21,1533.34,2.92,10.65
23,373.27,129.94,1031.32,1534.53,2.92,10.66
24,373.82,127.26,1033.87,1534.95,2.92,10.66
25,373.53,127.35,1042.13,1543.01,2.94,10.72
26,378.11,127.20,1054.10,1559.40,2.97,10.83
27,373.96,130.92,1003.64,1508.51,2.87,10.48
28,370.58,127.62,1047.41,1545.61,2.94,10.73
29,370.72,127.10,1049.21,1547.03,2.94,10.74
30,372.50,129.44,1051.85,1553.78,2.96,10.79
31,379.59,126.47,1049.45,1555.51,2.96,10.80
profile_csv=sevenb-ffn-tuning/fp32-b32.csv
