# MemVanta trained-model layer profile
prefill_ms=35038.8 decode_total_ms=9645.31 sampling_ms=0.604059
prefill_cycles_per_token=629632283.53 decode_cycles_per_token=1386575120.56
qkv_projection_ms=9946.28 attention_output_projection_ms=3329.98 ffn_gemm_ms=26740.57 output_head_ms=3390.41
non_gemm_core_ms=1276.31 (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
0,314.86,101.64,837.24,1253.74,2.89,8.71
1,310.71,104.24,834.59,1249.54,2.88,8.68
2,310.94,104.30,831.77,1247.01,2.87,8.66
3,309.81,107.19,831.61,1248.61,2.88,8.67
4,309.31,104.01,828.88,1242.20,2.86,8.63
5,306.97,104.01,839.02,1250.00,2.88,8.68
6,309.39,102.14,833.91,1245.45,2.87,8.65
7,309.25,103.28,840.40,1252.93,2.89,8.70
8,310.86,103.32,837.52,1251.70,2.88,8.69
9,309.74,104.77,837.93,1252.44,2.89,8.70
10,309.62,104.17,832.02,1245.81,2.87,8.65
11,310.19,104.54,843.73,1258.46,2.90,8.74
12,310.90,104.68,836.30,1251.88,2.88,8.69
13,311.45,104.40,833.01,1248.86,2.88,8.67
14,310.39,103.10,840.54,1254.04,2.89,8.71
15,310.54,105.71,835.33,1251.59,2.88,8.69
16,310.30,104.43,836.07,1250.80,2.88,8.69
17,311.23,104.04,837.44,1252.71,2.89,8.70
18,313.36,103.49,832.11,1248.97,2.88,8.67
19,309.31,105.51,839.03,1253.86,2.89,8.71
20,312.77,104.77,838.64,1256.17,2.89,8.72
21,310.05,103.01,834.72,1247.78,2.87,8.67
22,309.64,104.61,836.62,1250.87,2.88,8.69
23,311.77,103.21,833.44,1248.42,2.88,8.67
24,310.50,103.17,838.60,1252.27,2.88,8.70
25,309.71,102.13,835.57,1247.41,2.87,8.66
26,310.09,103.25,834.94,1248.27,2.88,8.67
27,314.60,102.73,830.21,1247.54,2.87,8.66
28,313.48,102.74,837.27,1253.49,2.89,8.70
29,312.19,104.69,836.02,1252.90,2.89,8.70
30,309.95,104.44,834.56,1248.96,2.88,8.67
31,312.40,108.25,831.52,1252.16,2.88,8.70
profile_csv=sevenb-ffn-kernel-ab/optimized-3.csv
