What it does
It exposes a vLLM API and includes game-agent examples and reproducible benchmarks.
Benchmarks & research
jevfire batches many Jev-style decisions over one shared context on CUDA language models.
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
It exposes a vLLM API and includes game-agent examples and reproducible benchmarks.
You can use this approach to help video game figures pick their next moves. Start by writing down the exact choices your character can make, like changing lanes or speeding up. You can also write down simple sentences describing what is happening around them.
A developer can connect your game to an AI model running on a graphics card. The model scores only your allowed list of actions instead of writing out long sentences. It cannot invent new choices, but your game must still check safety rules like avoiding collisions.