# MemVanta trained-model layer profile
prefill_ms=40744.6 decode_total_ms=8820.89 sampling_ms=0.373578
prefill_cycles_per_token=778421556.72 decode_cycles_per_token=1348176280.06
qkv_projection_ms=10831.39 attention_output_projection_ms=3621.30 ffn_gemm_ms=30615.26 output_head_ms=3196.24
prefill_qkv_projection_ms=9443.18 prefill_attention_output_projection_ms=3142.47 prefill_ffn_gemm_ms=26843.47 prefill_output_head_ms=375.51 prefill_kernel_ms=39804.63
decode_qkv_projection_ms=1388.21 decode_attention_output_projection_ms=478.83 decode_ffn_gemm_ms=3771.78 decode_output_head_ms=2820.73 decode_kernel_ms=8459.56
non_gemm_core_ms=1300.95 (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
0,339.27,113.17,960.26,1412.70,2.93,9.81
1,339.70,112.56,947.02,1399.27,2.90,9.72
2,339.67,113.24,960.75,1413.66,2.93,9.82
3,339.94,116.17,953.60,1409.71,2.92,9.79
4,338.42,112.71,958.09,1409.21,2.92,9.79
5,337.53,113.98,957.51,1409.02,2.92,9.78
6,339.04,114.07,961.70,1414.81,2.93,9.83
7,339.68,112.47,953.27,1405.42,2.91,9.76
8,336.82,112.72,967.18,1416.72,2.94,9.84
9,339.07,112.94,960.85,1412.86,2.93,9.81
10,337.17,114.57,948.38,1400.13,2.90,9.72
11,338.20,113.65,968.85,1420.71,2.94,9.87
12,339.04,112.72,965.36,1417.11,2.94,9.84
13,337.70,112.90,959.01,1409.62,2.92,9.79
14,337.49,112.69,965.88,1416.07,2.93,9.83
15,339.37,112.93,948.69,1400.99,2.90,9.73
16,339.33,113.63,964.48,1417.44,2.94,9.84
17,338.25,112.38,952.24,1402.88,2.91,9.74
18,338.91,113.10,953.61,1405.61,2.91,9.76
19,339.52,112.83,951.74,1404.08,2.91,9.75
20,338.28,113.40,954.26,1405.94,2.91,9.76
21,338.61,112.62,959.66,1410.89,2.92,9.80
22,337.11,112.55,958.63,1408.28,2.92,9.78
23,338.49,113.45,957.43,1409.38,2.92,9.79
24,337.85,113.57,954.23,1405.64,2.91,9.76
25,336.61,112.93,955.35,1404.89,2.91,9.76
26,338.90,112.88,954.32,1406.10,2.91,9.76
27,339.70,112.85,959.57,1412.11,2.93,9.81
28,338.99,112.94,954.25,1406.17,2.91,9.77
29,337.18,113.46,945.89,1396.53,2.89,9.70
30,337.76,112.62,955.08,1405.46,2.91,9.76
31,337.79,112.60,948.11,1398.50,2.90,9.71
profile_csv=sevenb-ffn-tuning/fp32-b16.csv
