Advisor / Submit
Submit your benchmark
Real numbers from real machines make the recommendations better. Submissions are shown as unverified community data, are never used to compute the estimates, and are reviewed before being published.
1. Run llama-bench
shell
llama-bench -m /path/to/model.gguf -p 512 -n 128 -ngl 99 -fa 1 -r 3
Defaults: 512 prompt tokens (pp512), 128 generated tokens (tg128). Use -ngl 99 for a model that fits fully on your GPU or unified memory; if you offload, say so in the notes. Get the build with llama-bench --version. Close other GPU-heavy apps first. Report the average (the first number before ±).
No benchmark is verified by us. Moderation: format and plausibility checks only. Data sources