Contact

The best way to reach us is by email: [email protected].

Useful things to include

• A number looks wrong: the page link, the GPU and model, and what you see on your own machine. If you ran a benchmark, the tool (llama.cpp, Ollama or LM Studio), the quant and the context length help us compare like with like.
• A model is missing: the Hugging Face link to the model, and a GGUF version if one exists. We add models once llama.cpp supports them.
• A GPU or chip is missing: its exact name and memory size, ideally with a link to the manufacturer's spec page.

We read every message. This is a small independent project, so replies can take a few days.

How the numbers are calculated is explained on the How we calculate page, and how we handle email is on the privacy page.