# MemVanta trained-model layer profile
prefill_ms=41977.6 decode_total_ms=9781.51 sampling_ms=0.634507
prefill_cycles_per_token=916112068.22 decode_cycles_per_token=1707758231.75
qkv_projection_ms=11438.03 attention_output_projection_ms=3915.59 ffn_gemm_ms=31012.47 output_head_ms=3779.93
prefill_qkv_projection_ms=10024.66 prefill_attention_output_projection_ms=3341.39 prefill_ffn_gemm_ms=27210.39 prefill_output_head_ms=420.17 prefill_kernel_ms=40996.61
decode_qkv_projection_ms=1413.37 decode_attention_output_projection_ms=574.20 decode_ffn_gemm_ms=3802.08 decode_output_head_ms=3359.76 decode_kernel_ms=9149.41
non_gemm_core_ms=1612.45 (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
0,359.68,120.50,971.98,1452.15,2.90,10.08
1,355.77,121.24,969.08,1446.08,2.88,10.04
2,357.11,124.18,965.29,1446.58,2.88,10.05
3,355.59,120.35,964.52,1440.46,2.87,10.00
4,358.17,127.94,968.29,1454.41,2.90,10.10
5,357.04,122.18,967.71,1446.93,2.89,10.05
6,355.54,120.65,970.62,1446.81,2.89,10.05
7,359.24,124.03,970.17,1453.44,2.90,10.09
8,356.69,123.92,969.98,1450.59,2.89,10.07
9,357.83,120.91,975.34,1454.08,2.90,10.10
10,358.64,122.26,970.94,1451.84,2.90,10.08
11,357.10,121.28,970.43,1448.80,2.89,10.06
12,356.70,121.47,969.78,1447.95,2.89,10.06
13,356.84,120.35,964.75,1441.93,2.88,10.01
14,357.07,122.76,972.29,1452.12,2.90,10.08
15,355.64,122.59,964.23,1442.47,2.88,10.02
16,356.34,122.89,973.12,1452.35,2.90,10.09
17,357.44,123.60,972.70,1453.74,2.90,10.10
18,359.35,120.16,974.56,1454.07,2.90,10.10
19,355.72,120.86,965.69,1442.27,2.88,10.02
20,361.26,123.58,968.52,1453.36,2.90,10.09
21,356.62,122.77,970.03,1449.43,2.89,10.07
22,357.05,124.67,967.38,1449.09,2.89,10.06
23,358.35,121.45,968.32,1448.11,2.89,10.06
24,355.51,121.74,966.97,1444.22,2.88,10.03
25,356.82,123.32,971.59,1451.74,2.90,10.08
26,360.56,122.90,971.59,1455.06,2.90,10.10
27,357.88,122.12,968.93,1448.92,2.89,10.06
28,357.06,122.65,965.55,1445.26,2.88,10.04
29,357.85,123.28,966.69,1447.81,2.89,10.05
30,358.55,121.05,966.61,1446.20,2.88,10.04
31,357.04,121.94,968.83,1447.81,2.89,10.05
profile_csv=sevenb-throughput-profile/layer-profile.csv
