All tracers
llama.cpp
inference tracer ·
2ca15f5
· Llama 3.1 8B Q4_K_M · RTX 4090
Decode step
Prompt pass
Layer
−
100%
+
Fit
Data path of one forward pass
Repeats: next layer, next token
Last layer only
⏮
◀
▶ Trace
▶▶
TRACE A FORWARD PASS
Press ▶ to follow one token through the model
Follow