glasstrace report

Per-layer inference profile · device: cpu

Total time
166.4 ms
Prefill
17.9 ms
Decode total
148.5 ms
ms / token
14.8
Modules
169
KV-cache growth
0.00 MB

Top 30 modules by total time

prefill decode

All modules — top 30 by total time

Module Type Phase Calls Total ms ↓ Per-call ms % of total
lm_head Linear decode 10 37.01 3.70 22.2%
model.layers.6.mlp.down_proj Linear decode 10 1.52 0.15 0.9%
model.layers.10.mlp.down_proj Linear decode 10 1.51 0.15 0.9%
model.layers.6.mlp.up_proj Linear decode 10 1.50 0.15 0.9%
model.layers.10.mlp.up_proj Linear decode 10 1.49 0.15 0.9%
model.layers.12.mlp.down_proj Linear decode 10 1.49 0.15 0.9%
model.layers.9.mlp.up_proj Linear decode 10 1.48 0.15 0.9%
model.layers.0.mlp.up_proj Linear decode 10 1.47 0.15 0.9%
model.layers.15.mlp.up_proj Linear decode 10 1.45 0.15 0.9%
model.layers.13.mlp.down_proj Linear decode 10 1.45 0.15 0.9%
model.layers.7.mlp.up_proj Linear decode 10 1.45 0.14 0.9%
model.layers.9.mlp.down_proj Linear decode 10 1.45 0.14 0.9%
model.layers.3.mlp.up_proj Linear decode 10 1.44 0.14 0.9%
model.layers.8.mlp.down_proj Linear decode 10 1.44 0.14 0.9%
model.layers.0.mlp.gate_proj Linear decode 10 1.44 0.14 0.9%
model.layers.11.mlp.down_proj Linear decode 10 1.44 0.14 0.9%
model.layers.12.mlp.gate_proj Linear decode 10 1.44 0.14 0.9%
model.layers.6.mlp.gate_proj Linear decode 10 1.44 0.14 0.9%
model.layers.1.mlp.up_proj Linear decode 10 1.43 0.14 0.9%
model.layers.9.mlp.gate_proj Linear decode 10 1.43 0.14 0.9%
model.layers.5.mlp.up_proj Linear decode 10 1.43 0.14 0.9%
model.layers.11.mlp.gate_proj Linear decode 10 1.43 0.14 0.9%
model.layers.17.mlp.down_proj Linear decode 10 1.43 0.14 0.9%
model.layers.2.mlp.down_proj Linear decode 10 1.43 0.14 0.9%
model.layers.5.mlp.down_proj Linear decode 10 1.43 0.14 0.9%
model.layers.18.mlp.down_proj Linear decode 10 1.43 0.14 0.9%
model.layers.8.mlp.up_proj Linear decode 10 1.42 0.14 0.9%
model.layers.2.mlp.up_proj Linear decode 10 1.42 0.14 0.9%
model.layers.15.mlp.down_proj Linear decode 10 1.42 0.14 0.9%
model.layers.10.mlp.gate_proj Linear decode 10 1.42 0.14 0.9%