# MemVanta trained-model layer profile
prefill_ms=51505.9 decode_total_ms=8966.68 sampling_ms=0.36915
prefill_cycles_per_token=984014103.11 decode_cycles_per_token=1370458604.38
qkv_projection_ms=13531.06 attention_output_projection_ms=4564.53 ffn_gemm_ms=37905.81 output_head_ms=3240.31
prefill_qkv_projection_ms=12112.29 prefill_attention_output_projection_ms=4070.27 prefill_ffn_gemm_ms=34071.96 prefill_output_head_ms=375.82 prefill_kernel_ms=50630.33
decode_qkv_projection_ms=1418.77 decode_attention_output_projection_ms=494.26 decode_ffn_gemm_ms=3833.85 decode_output_head_ms=2864.49 decode_kernel_ms=8611.37
non_gemm_core_ms=1230.47 (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
0,428.27,141.72,1198.34,1768.34,2.98,12.28
1,431.13,141.45,1226.44,1799.02,3.04,12.49
2,425.34,145.23,1182.92,1753.49,2.96,12.18
3,409.26,142.40,1192.81,1744.46,2.94,12.11
4,419.81,142.43,1178.94,1741.18,2.94,12.09
5,422.70,142.22,1205.93,1770.85,2.99,12.30
6,419.02,139.70,1149.86,1708.58,2.88,11.87
7,431.10,139.22,1165.52,1735.84,2.93,12.05
8,405.62,139.94,1180.06,1725.62,2.91,11.98
9,431.34,144.88,1188.20,1764.42,2.98,12.25
10,424.63,142.37,1201.06,1768.06,2.98,12.28
11,423.08,139.77,1204.60,1767.45,2.98,12.27
12,423.42,149.61,1206.85,1779.88,3.00,12.36
13,421.70,141.16,1198.90,1761.75,2.97,12.23
14,419.36,139.94,1172.09,1731.39,2.92,12.02
15,422.73,140.28,1177.84,1740.85,2.94,12.09
16,427.21,140.67,1196.07,1763.96,2.98,12.25
17,416.80,139.71,1167.46,1723.97,2.91,11.97
18,427.70,144.47,1189.43,1761.60,2.97,12.23
19,433.31,143.68,1129.58,1706.57,2.88,11.85
20,420.15,148.60,1204.30,1773.05,2.99,12.31
21,418.35,145.32,1177.69,1741.36,2.94,12.09
22,412.46,144.93,1173.45,1730.84,2.92,12.02
23,424.94,134.93,1167.50,1727.37,2.92,12.00
24,404.61,145.64,1202.37,1752.62,2.96,12.17
25,435.34,137.69,1196.37,1769.40,2.99,12.29
26,418.44,141.91,1187.17,1747.53,2.95,12.14
27,430.33,146.48,1186.21,1763.02,2.98,12.24
28,422.85,144.71,1135.62,1703.19,2.87,11.83
29,429.69,142.65,1174.52,1746.86,2.95,12.13
30,423.78,144.06,1200.38,1768.22,2.98,12.28
31,426.62,146.75,1187.31,1760.68,2.97,12.23
profile_csv=sevenb-ffn-tuning/fp32-b64.csv
