# MemVanta trained-model layer profile
prefill_ms=35219.5 decode_total_ms=9661.81 sampling_ms=0.634397
prefill_cycles_per_token=632878294.22 decode_cycles_per_token=1388960582.94
qkv_projection_ms=9978.63 attention_output_projection_ms=3334.05 ffn_gemm_ms=26871.69 output_head_ms=3392.45
non_gemm_core_ms=1303.86 (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
0,312.89,103.12,840.60,1256.60,2.88,8.73
1,312.91,104.04,836.66,1253.61,2.88,8.71
2,311.84,104.23,840.71,1256.77,2.88,8.73
3,311.91,103.19,834.32,1249.42,2.87,8.68
4,310.96,105.60,838.52,1255.08,2.88,8.72
5,312.44,103.90,836.53,1252.88,2.88,8.70
6,313.95,104.25,840.06,1258.25,2.89,8.74
7,310.37,103.72,839.05,1253.14,2.88,8.70
8,311.22,103.51,840.80,1255.53,2.88,8.72
9,310.64,103.97,842.86,1257.47,2.89,8.73
10,311.58,103.48,841.16,1256.22,2.88,8.72
11,312.05,103.47,839.28,1254.79,2.88,8.71
12,314.99,103.59,841.54,1260.12,2.89,8.75
13,313.03,104.62,842.11,1259.76,2.89,8.75
14,311.05,104.18,837.62,1252.84,2.88,8.70
15,312.00,104.25,839.04,1255.30,2.88,8.72
16,313.99,103.57,840.18,1257.74,2.89,8.73
17,311.32,105.36,839.56,1256.24,2.88,8.72
18,312.17,104.67,840.23,1257.07,2.88,8.73
19,312.27,104.43,837.32,1254.02,2.88,8.71
20,310.12,103.61,840.29,1254.02,2.88,8.71
21,311.83,104.40,837.73,1253.96,2.88,8.71
22,310.81,104.00,839.86,1254.67,2.88,8.71
23,312.32,104.21,841.60,1258.14,2.89,8.74
24,313.82,104.46,840.63,1258.92,2.89,8.74
25,312.02,104.21,839.40,1255.63,2.88,8.72
26,311.67,104.40,843.30,1259.37,2.89,8.75
27,308.00,104.69,838.95,1251.64,2.87,8.69
28,313.64,104.15,840.50,1258.29,2.89,8.74
29,310.18,104.48,838.94,1253.60,2.88,8.71
30,310.30,105.13,841.55,1256.99,2.88,8.73
31,310.37,105.13,840.79,1256.29,2.88,8.72
profile_csv=sevenb-ffn-kernel-ab/optimized-2.csv
