# MemVanta trained-model layer profile
prefill_ms=90925.4 decode_total_ms=8786.65 sampling_ms=0.363118
prefill_cycles_per_token=1737120681.22 decode_cycles_per_token=1342943980.00
qkv_projection_ms=23702.49 attention_output_projection_ms=7910.98 ffn_gemm_ms=63623.23 output_head_ms=3197.30
prefill_qkv_projection_ms=22323.50 prefill_attention_output_projection_ms=7438.36 prefill_ffn_gemm_ms=59861.87 prefill_output_head_ms=375.23 prefill_kernel_ms=89998.97
decode_qkv_projection_ms=1378.99 decode_attention_output_projection_ms=472.61 decode_ffn_gemm_ms=3761.36 decode_output_head_ms=2822.07 decode_kernel_ms=8435.03
non_gemm_core_ms=1277.72 (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
0,743.00,248.46,1988.45,2979.91,3.03,20.69
1,741.16,247.08,1987.98,2976.22,3.02,20.67
2,739.80,247.26,1987.98,2975.05,3.02,20.66
3,741.06,247.46,1989.62,2978.15,3.03,20.68
4,741.19,247.85,1988.48,2977.52,3.02,20.68
5,740.41,246.91,1987.70,2975.02,3.02,20.66
6,740.89,246.68,1989.79,2977.37,3.02,20.68
7,741.21,247.01,1987.78,2975.99,3.02,20.67
8,740.84,247.38,1987.57,2975.79,3.02,20.67
9,740.53,247.23,1988.36,2976.12,3.02,20.67
10,740.04,247.74,1989.38,2977.16,3.02,20.67
11,741.15,247.13,1987.57,2975.85,3.02,20.67
12,740.69,247.12,1986.70,2974.50,3.02,20.66
13,740.06,247.28,1985.83,2973.17,3.02,20.65
14,740.76,246.81,1988.54,2976.11,3.02,20.67
15,739.92,247.39,1988.06,2975.36,3.02,20.66
16,740.93,247.51,1988.25,2976.69,3.02,20.67
17,743.11,247.54,1988.08,2978.74,3.03,20.69
18,739.83,247.17,1986.40,2973.40,3.02,20.65
19,740.63,246.94,1988.80,2976.37,3.02,20.67
20,739.82,247.32,1988.10,2975.23,3.02,20.66
21,740.27,247.56,1987.26,2975.09,3.02,20.66
22,740.43,247.33,1987.34,2975.10,3.02,20.66
23,740.36,246.89,1989.55,2976.80,3.02,20.67
24,740.55,247.03,1989.49,2977.07,3.02,20.67
25,740.63,247.66,1987.53,2975.82,3.02,20.67
26,741.80,246.80,1987.67,2976.26,3.02,20.67
27,740.17,246.47,1988.64,2975.29,3.02,20.66
28,740.14,247.02,1988.46,2975.62,3.02,20.66
29,739.74,246.96,1988.77,2975.47,3.02,20.66
30,740.83,246.94,1988.65,2976.42,3.02,20.67
31,740.54,247.03,1990.47,2978.05,3.03,20.68
profile_csv=sevenb-ffn-tuning/q8act-b16.csv
