# MemVanta trained-model layer profile
prefill_ms=90685.9 decode_total_ms=8810.94 sampling_ms=0.363009
prefill_cycles_per_token=1732544805.01 decode_cycles_per_token=1346655537.06
qkv_projection_ms=23682.54 attention_output_projection_ms=7910.05 ffn_gemm_ms=63494.41 output_head_ms=3196.65
prefill_qkv_projection_ms=22289.46 prefill_attention_output_projection_ms=7431.32 prefill_ffn_gemm_ms=59717.74 prefill_output_head_ms=375.52 prefill_kernel_ms=89814.04
decode_qkv_projection_ms=1393.08 decode_attention_output_projection_ms=478.74 decode_ffn_gemm_ms=3776.67 decode_output_head_ms=2821.13 decode_kernel_ms=8469.62
non_gemm_core_ms=1212.80 (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
0,741.52,246.85,1986.95,2975.33,3.03,20.66
1,740.51,247.66,1985.09,2973.26,3.03,20.65
2,739.68,246.84,1983.69,2970.21,3.02,20.63
3,743.01,247.04,1985.78,2975.83,3.03,20.67
4,740.04,247.04,1984.53,2971.60,3.02,20.64
5,739.98,247.25,1988.32,2975.56,3.03,20.66
6,743.04,249.79,1987.51,2980.34,3.03,20.70
7,740.02,246.96,1984.19,2971.18,3.02,20.63
8,740.43,246.93,1986.52,2973.87,3.03,20.65
9,739.55,247.29,1982.91,2969.75,3.02,20.62
10,739.49,246.83,1982.46,2968.78,3.02,20.62
11,739.08,247.09,1982.93,2969.11,3.02,20.62
12,739.46,246.89,1982.40,2968.75,3.02,20.62
13,742.77,247.09,1985.71,2975.57,3.03,20.66
14,743.09,247.41,1985.59,2976.09,3.03,20.67
15,739.66,246.87,1983.12,2969.65,3.02,20.62
16,739.35,246.73,1983.61,2969.68,3.02,20.62
17,739.47,246.88,1982.72,2969.08,3.02,20.62
18,739.56,246.79,1982.66,2969.01,3.02,20.62
19,739.26,247.03,1983.57,2969.86,3.02,20.62
20,739.46,247.05,1983.39,2969.89,3.02,20.62
21,739.12,246.99,1983.35,2969.47,3.02,20.62
22,739.49,247.09,1983.00,2969.58,3.02,20.62
23,739.22,247.22,1984.29,2970.73,3.02,20.63
24,739.95,247.06,1983.72,2970.73,3.02,20.63
25,739.44,247.47,1983.29,2970.21,3.02,20.63
26,739.53,248.56,1984.64,2972.74,3.02,20.64
27,739.37,246.72,1983.03,2969.11,3.02,20.62
28,739.53,247.53,1983.58,2970.65,3.02,20.63
29,739.36,246.82,1983.07,2969.25,3.02,20.62
30,739.09,247.70,1984.19,2970.98,3.02,20.63
31,739.98,246.57,1984.60,2971.15,3.02,20.63
profile_csv=sevenb-ffn-tuning/q8act-b32.csv
