Tiny AI Models Dashboard

Visualize and compare evaluation results across different model runs

This is the working companion to The Unreasonable Effectiveness of Tiny AI Models. The project evaluates small language models on standard benchmarks such as HellaSwag and MMLU, then reruns those tests with Retrieval-Augmented Generation from Simple Wikipedia so you can see whether a pocket-sized model gets smarter when it can look things up. Pick a model and evaluation run below to inspect accuracy, subject breakdowns, and detailed scores.