Local AI Speedrun: Biggest Model, Smallest Box

​No cloud. No API keys. Just you, your hardware and the biggest model you can make it run.

​A one-day hack for people who run AI locally. Bring a laptop, a mini PC or a GPU rig, and race to squeeze the most out of open-weight models. A live leaderboard runs all afternoon.

​Three races

• ​Biggest model: the largest model you can run at 5+ tokens a second on hardware you carried in

• ​Fastest tokens: highest tokens per second on a fixed model and prompt set we hand out on the day

• ​Most useful offline app: build something genuinely useful that works with the Wi-Fi switched off

​The day

• ​12:00pm: doors, pizza, rules and the leaderboard goes live

• ​12:30pm: hacking (quantisation, runtimes, tricks: anything goes as long as it runs on the box in front of you)

• ​5:00pm: final runs, verified live

• ​5:30pm: winners and drinks

​Prizes for each race, and the top prize is a local-AI mini PC to keep.

​Beginners welcome: if you've never run a model locally, we'll get you going in ten minutes with Ollama. Hosted by Encode Club at Encode Hub, Shoreditch.