# MemVanta trained-model layer profile
prefill_ms=43933.8 decode_total_ms=8791.88 sampling_ms=0.355315
prefill_cycles_per_token=839342978.60 decode_cycles_per_token=1343732013.31
qkv_projection_ms=11758.24 attention_output_projection_ms=4139.92 ffn_gemm_ms=32416.41 output_head_ms=3196.74
prefill_qkv_projection_ms=10372.36 prefill_attention_output_projection_ms=3665.62 prefill_ffn_gemm_ms=28652.57 prefill_output_head_ms=375.44 prefill_kernel_ms=43066.00
decode_qkv_projection_ms=1385.88 decode_attention_output_projection_ms=474.29 decode_ffn_gemm_ms=3763.84 decode_output_head_ms=2821.30 decode_kernel_ms=8445.31
non_gemm_core_ms=1214.00 (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
0,373.69,127.20,1015.58,1516.48,2.94,10.53
1,368.99,131.13,1011.44,1511.55,2.93,10.50
2,365.75,130.92,1011.15,1507.81,2.93,10.47
3,356.15,127.94,998.98,1483.07,2.88,10.30
4,354.55,129.43,1022.46,1506.43,2.92,10.46
5,360.36,129.25,1007.71,1497.32,2.91,10.40
6,366.60,130.58,1014.93,1512.10,2.94,10.50
7,365.73,130.53,1020.92,1517.18,2.95,10.54
8,369.12,130.41,1011.80,1511.33,2.93,10.50
9,368.57,127.71,1022.59,1518.86,2.95,10.55
10,363.16,127.88,1021.68,1512.71,2.94,10.50
11,364.18,130.21,1033.98,1528.37,2.97,10.61
12,378.58,127.19,1010.35,1516.11,2.94,10.53
13,363.12,130.61,1015.87,1509.60,2.93,10.48
14,374.93,128.99,1022.48,1526.40,2.96,10.60
15,373.65,131.25,1022.45,1527.35,2.97,10.61
16,369.81,128.24,981.29,1479.34,2.87,10.27
17,360.24,127.47,1007.34,1495.05,2.90,10.38
18,376.36,131.62,1025.50,1533.48,2.98,10.65
19,361.90,127.53,1004.34,1493.77,2.90,10.37
20,358.98,128.74,1021.86,1509.59,2.93,10.48
21,370.43,129.52,996.73,1496.69,2.91,10.39
22,365.48,128.09,1026.64,1520.21,2.95,10.56
23,368.49,130.28,1015.05,1513.82,2.94,10.51
24,372.59,129.61,1011.46,1513.65,2.94,10.51
25,372.70,130.79,1009.07,1512.56,2.94,10.50
26,371.41,127.69,983.71,1482.80,2.88,10.30
27,369.56,126.70,1020.53,1516.79,2.94,10.53
28,361.32,131.47,1013.33,1506.13,2.92,10.46
29,373.33,129.37,1035.13,1537.82,2.99,10.68
30,371.68,130.34,1015.75,1517.76,2.95,10.54
31,366.85,131.24,984.31,1482.40,2.88,10.29
profile_csv=sevenb-ffn-kernel-ab/optimized-1.csv
