Playground

No hosted endpoint is configured on this site. You can connect the playground to a Mtrini model running on your own machine, or use demo mode, which shows the request that would be sent without calling a model.

Demo mode · simulated

Try the interface.

Demo mode never calls a model. Send a prompt to see the exact request a configured deployment would make.

Example prompts from the model cards

Running a model locally

Mtrini GGUF builds run with llama.cpp (opens in new tab) and other GGUF runtimes. Start an OpenAI-compatible server, then pick Local endpoint above. Your browser sends requests straight to that server.

Hosting the playground

Site operators turn on hosted mode with server-side environment variables: PLAYGROUND_API_BASE_URL, PLAYGROUND_API_KEY and PLAYGROUND_MODELS. See .env.example in the site repository. Keys stay on the server, and requests are validated and rate-limited.

Local quick start

# 1. Download a Mtrini GGUF from Hugging Face, e.g.
#    huggingface.co/CompiwerAI/Mtrini-SVL-1.0-GGUF  (Mtrini-SVL-1.0-Q8_0.gguf, ~8.7 GB)

# 2. Serve it with llama.cpp's OpenAI-compatible server
llama-server -m Mtrini-SVL-1.0-Q8_0.gguf --port 8080

# 3. In Local endpoint mode use:
#    URL       http://127.0.0.1:8080/v1
#    Model ID  the id listed at http://127.0.0.1:8080/v1/models

Commands follow the model cards and llama.cpp’s documentation. Flags may differ between llama.cpp versions.