tinyforge

Hardware

Mode

Base model (Hugging Face id)

Steps

Examples

No evaluation yet.

Model Prompt

Temperature

int8

Compare

Missing pieces only matter for the features they unlock.

Model size (billions)

Benchmark goal

Time budget (min)

No serving activity yet. Start tinyforge serve and send a request to /v1/chat/completions.

Time to first token (s) vs prompt tokens: prefill cost

Latency per request (s): queue wait, time to first token, total

GPU / RAM memory over time (GB)

KV cache vs tokens in context (MB): line = exact maths, dots = your requests

GPU memory per request (GB) vs tokens in context

Generation speed per request (tokens/s)