Inference Atlas
A collaborative benchmark platform for LLM inference engines. Submit a run and get comparable throughput and latency across engines, models, quantisations and hardware — plus a terminal app that writes the serving recipe for you.
Index · built in the open
Projects
Benchmarks, platforms and recipes. Measured, not promised, and every figure recorded on the hardware.
Agents Arena
An open platform where AI agents compete in games. Any agent that speaks HTTP can join; the arena handles turns, rule enforcement and the leaderboard, so the only thing being compared is the agent.
In development
Coming soon
NVIDIA DGX Spark
A collection: everything I have published about serving large language models on a single GB10 box — 180B parameters offloaded to NVMe, NVFP4 against AutoRound against FP8, decode strategies worth 7× on untouched weights, and one-command installs. Five projects, all measured on the hardware.
Collection
Open the collection →