# MemVanta trained-model layer profile
prefill_ms=30564.5 decode_total_ms=9645.85 sampling_ms=0.620065
prefill_cycles_per_token=549231229.97 decode_cycles_per_token=1386653150.06
qkv_projection_ms=8933.35 attention_output_projection_ms=2973.92 ffn_gemm_ms=23826.93 output_head_ms=3394.50
non_gemm_core_ms=1081.05 (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
0,280.18,93.77,743.89,1117.83,2.86,7.76
1,281.44,92.93,743.34,1117.71,2.86,7.76
2,279.61,93.25,745.25,1118.11,2.86,7.76
3,279.93,92.26,744.08,1116.26,2.85,7.75
4,279.23,93.13,745.07,1117.44,2.86,7.76
5,280.04,92.21,746.49,1118.74,2.86,7.77
6,278.88,93.14,745.72,1117.74,2.86,7.76
7,278.27,92.93,744.59,1115.78,2.85,7.75
8,279.30,92.82,744.70,1116.83,2.85,7.76
9,279.22,92.72,743.37,1115.31,2.85,7.75
10,278.89,92.32,740.85,1112.06,2.84,7.72
11,279.64,92.65,744.12,1116.40,2.85,7.75
12,279.32,93.06,745.47,1117.85,2.86,7.76
13,278.91,92.76,743.99,1115.66,2.85,7.75
14,277.42,93.09,744.40,1114.91,2.85,7.74
15,280.37,92.71,745.80,1118.88,2.86,7.77
16,279.23,93.85,743.35,1116.44,2.85,7.75
17,279.64,92.58,744.84,1117.06,2.85,7.76
18,278.89,92.48,744.50,1115.87,2.85,7.75
19,279.62,93.57,745.11,1118.30,2.86,7.77
20,278.14,92.63,745.29,1116.06,2.85,7.75
21,277.47,91.77,743.90,1113.13,2.84,7.73
22,278.36,93.25,745.06,1116.68,2.85,7.75
23,277.72,93.43,745.75,1116.90,2.85,7.76
24,278.80,93.28,745.18,1117.26,2.86,7.76
25,280.67,92.00,744.02,1116.69,2.85,7.75
26,278.53,93.97,743.16,1115.66,2.85,7.75
27,280.03,93.21,744.02,1117.26,2.86,7.76
28,281.29,93.25,746.68,1121.22,2.87,7.79
29,278.46,92.45,744.96,1115.88,2.85,7.75
30,278.31,94.17,743.44,1115.92,2.85,7.75
31,277.56,92.27,746.54,1116.37,2.85,7.75
profile_csv=sevenb-ffn-kernel-ab/baseline-2.csv
