# MemVanta trained-model layer profile
prefill_ms=35341.8 decode_total_ms=9673.91 sampling_ms=0.610864
prefill_cycles_per_token=635075132.84 decode_cycles_per_token=1390685827.12
qkv_projection_ms=9997.88 attention_output_projection_ms=3345.93 ffn_gemm_ms=26969.18 output_head_ms=3391.35
non_gemm_core_ms=1310.75 (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
0,310.43,104.53,841.28,1256.25,2.87,8.72
1,312.54,105.23,844.53,1262.30,2.89,8.77
2,317.70,102.99,843.57,1264.26,2.89,8.78
3,310.63,105.70,843.35,1259.68,2.88,8.75
4,311.95,106.38,841.72,1260.05,2.88,8.75
5,316.79,104.06,845.03,1265.87,2.90,8.79
6,312.07,103.66,841.81,1257.54,2.88,8.73
7,310.64,104.83,842.10,1257.57,2.88,8.73
8,309.45,105.33,841.33,1256.11,2.87,8.72
9,313.43,104.66,840.80,1258.89,2.88,8.74
10,310.32,104.34,840.86,1255.52,2.87,8.72
11,311.42,104.68,840.33,1256.43,2.87,8.73
12,316.16,105.95,842.44,1264.54,2.89,8.78
13,312.38,104.73,841.05,1258.16,2.88,8.74
14,313.27,104.93,841.67,1259.87,2.88,8.75
15,312.90,104.12,842.31,1259.33,2.88,8.75
16,313.37,102.54,844.03,1259.94,2.88,8.75
17,312.66,104.81,845.63,1263.10,2.89,8.77
18,312.98,104.15,845.71,1262.84,2.89,8.77
19,311.52,103.86,841.75,1257.13,2.88,8.73
20,311.63,104.54,842.69,1258.86,2.88,8.74
21,311.47,103.60,844.88,1259.96,2.88,8.75
22,312.97,107.36,847.25,1267.59,2.90,8.80
23,314.31,104.45,846.83,1265.59,2.90,8.79
24,312.15,104.52,841.08,1257.75,2.88,8.73
25,310.19,104.97,840.54,1255.70,2.87,8.72
26,312.95,106.14,841.81,1260.91,2.89,8.76
27,311.73,105.65,843.99,1261.37,2.89,8.76
28,311.37,103.01,841.80,1256.18,2.87,8.72
29,312.09,103.87,841.96,1257.91,2.88,8.74
30,311.14,103.30,843.83,1258.27,2.88,8.74
31,313.27,103.02,841.22,1257.51,2.88,8.73
profile_csv=sevenb-ffn-kernel-ab/optimized-1.csv
