# MemVanta trained-model layer profile
prefill_ms=35184.7 decode_total_ms=9652.32 sampling_ms=0.615191
prefill_cycles_per_token=632252005.78 decode_cycles_per_token=1387579948.75
qkv_projection_ms=10002.93 attention_output_projection_ms=3355.14 ffn_gemm_ms=26998.12 output_head_ms=3390.37
non_gemm_core_ms=1089.83 (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
0,315.38,105.84,844.22,1265.44,2.89,8.79
1,315.97,105.18,845.33,1266.47,2.90,8.79
2,312.10,105.11,842.17,1259.38,2.88,8.75
3,312.86,103.20,844.85,1260.91,2.88,8.76
4,311.57,104.82,843.35,1259.74,2.88,8.75
5,311.48,105.17,844.55,1261.19,2.88,8.76
6,312.00,104.28,841.47,1257.75,2.88,8.73
7,316.48,105.54,843.70,1265.73,2.89,8.79
8,311.81,107.61,840.27,1259.69,2.88,8.75
9,312.91,104.26,842.16,1259.33,2.88,8.75
10,311.50,106.35,841.19,1259.04,2.88,8.74
11,311.14,103.45,842.18,1256.77,2.87,8.73
12,313.99,104.46,844.07,1262.52,2.89,8.77
13,310.38,104.62,845.97,1260.96,2.88,8.76
14,312.33,103.69,847.37,1263.39,2.89,8.77
15,312.18,105.50,849.15,1266.83,2.90,8.80
16,310.64,102.96,843.37,1256.97,2.87,8.73
17,309.98,105.09,846.54,1261.61,2.88,8.76
18,312.87,103.11,845.00,1260.98,2.88,8.76
19,312.45,104.92,844.04,1261.41,2.88,8.76
20,312.73,105.90,842.19,1260.83,2.88,8.76
21,313.08,105.89,846.03,1265.00,2.89,8.78
22,313.70,103.73,844.50,1261.93,2.88,8.76
23,313.01,105.88,843.14,1262.04,2.88,8.76
24,311.53,104.91,843.45,1259.88,2.88,8.75
25,312.44,105.10,838.73,1256.27,2.87,8.72
26,313.34,103.42,842.65,1259.41,2.88,8.75
27,312.51,104.41,843.61,1260.53,2.88,8.75
28,315.19,105.95,840.89,1262.02,2.88,8.76
29,310.60,104.77,843.61,1258.98,2.88,8.74
30,312.78,105.44,845.35,1263.57,2.89,8.77
31,311.98,104.59,843.01,1259.57,2.88,8.75
profile_csv=sevenb-ffn-kernel-ab/baseline-1.csv
