# MemVanta trained-model layer profile
prefill_ms=45402.4 decode_total_ms=8904.9 sampling_ms=0.358635
prefill_cycles_per_token=867401586.73 decode_cycles_per_token=1361005393.81
qkv_projection_ms=12202.69 attention_output_projection_ms=4117.15 ffn_gemm_ms=33506.94 output_head_ms=3264.41
prefill_qkv_projection_ms=10801.06 prefill_attention_output_projection_ms=3637.48 prefill_ffn_gemm_ms=29716.54 prefill_output_head_ms=375.43 prefill_kernel_ms=44530.51
decode_qkv_projection_ms=1401.63 decode_attention_output_projection_ms=479.67 decode_ffn_gemm_ms=3790.40 decode_output_head_ms=2888.98 decode_kernel_ms=8560.69
non_gemm_core_ms=1215.78 (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
0,395.08,132.55,1073.39,1601.02,3.02,11.12
1,385.60,127.84,1066.04,1579.48,2.98,10.97
2,384.36,128.98,1033.56,1546.91,2.91,10.74
3,382.69,130.09,1040.52,1553.30,2.93,10.79
4,377.27,129.17,1070.05,1576.50,2.97,10.95
5,385.29,127.76,1057.48,1570.53,2.96,10.91
6,383.66,127.96,1050.85,1562.47,2.94,10.85
7,382.63,129.03,1050.11,1561.77,2.94,10.85
8,383.03,127.37,1042.77,1553.17,2.93,10.79
9,380.73,129.13,1037.75,1547.60,2.91,10.75
10,380.63,128.26,1039.50,1548.38,2.92,10.75
11,379.51,127.45,1035.19,1542.15,2.90,10.71
12,383.79,129.92,1062.69,1576.39,2.97,10.95
13,379.61,127.88,1068.49,1575.98,2.97,10.94
14,379.88,127.78,1065.58,1573.24,2.96,10.93
15,377.65,129.10,1057.68,1564.43,2.95,10.86
16,381.53,129.62,1053.30,1564.45,2.95,10.86
17,376.47,129.07,1032.57,1538.12,2.90,10.68
18,378.19,129.61,1038.05,1545.84,2.91,10.74
19,383.33,130.41,1049.51,1563.24,2.94,10.86
20,380.61,128.10,1054.26,1562.97,2.94,10.85
21,379.18,129.34,1043.47,1551.99,2.92,10.78
22,380.43,127.06,1023.32,1530.82,2.88,10.63
23,381.96,129.42,1018.52,1529.91,2.88,10.62
24,382.93,129.25,1052.68,1564.85,2.95,10.87
25,384.20,129.07,1042.20,1555.47,2.93,10.80
26,380.65,128.05,1057.78,1566.47,2.95,10.88
27,379.85,128.25,1029.15,1537.25,2.90,10.68
28,378.44,127.55,1027.76,1533.75,2.89,10.65
29,379.07,127.44,1052.39,1558.90,2.94,10.83
30,376.29,128.19,1039.89,1544.37,2.91,10.72
31,378.18,126.47,1040.44,1545.09,2.91,10.73
profile_csv=sevenb-ffn-kernel-ab/optimized-3.csv
