# MemVanta trained-model layer profile
prefill_ms=30601.7 decode_total_ms=9675.76 sampling_ms=0.600359
prefill_cycles_per_token=549899468.94 decode_cycles_per_token=1390953133.62
qkv_projection_ms=8942.05 attention_output_projection_ms=2992.90 ffn_gemm_ms=23882.33 output_head_ms=3393.20
non_gemm_core_ms=1066.41 (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
0,281.44,93.09,744.90,1119.43,2.85,7.77
1,279.06,92.66,744.67,1116.39,2.85,7.75
2,280.90,93.51,745.47,1119.87,2.86,7.78
3,278.88,94.21,745.09,1118.18,2.85,7.77
4,279.36,93.54,744.83,1117.72,2.85,7.76
5,280.85,93.16,746.05,1120.06,2.86,7.78
6,279.84,92.92,745.26,1118.02,2.85,7.76
7,279.10,93.92,745.64,1118.65,2.85,7.77
8,277.47,92.49,743.99,1113.96,2.84,7.74
9,277.45,93.72,745.96,1117.13,2.85,7.76
10,278.06,93.44,744.87,1116.37,2.85,7.75
11,277.91,93.44,746.11,1117.47,2.85,7.76
12,278.91,93.29,745.70,1117.91,2.85,7.76
13,279.33,93.63,744.61,1117.57,2.85,7.76
14,278.14,93.39,747.22,1118.75,2.85,7.77
15,280.77,92.83,746.05,1119.66,2.86,7.78
16,279.38,93.33,745.92,1118.63,2.85,7.77
17,279.30,92.31,745.52,1117.13,2.85,7.76
18,279.08,92.80,743.70,1115.58,2.85,7.75
19,279.64,92.77,744.43,1116.84,2.85,7.76
20,278.47,93.23,745.48,1117.17,2.85,7.76
21,279.14,94.69,744.57,1118.40,2.85,7.77
22,277.61,93.27,746.59,1117.47,2.85,7.76
23,279.35,97.78,751.64,1128.76,2.88,7.84
24,279.81,95.44,753.13,1128.39,2.88,7.84
25,281.42,96.34,751.01,1128.77,2.88,7.84
26,282.57,92.85,751.47,1126.90,2.87,7.83
27,279.24,92.87,752.36,1124.48,2.87,7.81
28,283.24,92.94,744.26,1120.45,2.86,7.78
29,279.33,93.13,744.47,1116.94,2.85,7.76
30,278.76,92.64,746.14,1117.54,2.85,7.76
31,278.21,93.25,745.21,1116.68,2.85,7.75
profile_csv=sevenb-ffn-kernel-ab/baseline-3.csv
