MicroLLM Lab allows in-browser testing of seven small LLMs

In-Browser LLM Testing

MicroLLM Lab offers an environment where users can interact with and test seven different small language models (LLMs) directly within their web browser. This platform facilitates immediate evaluation without requiring any local installation or complex setup for the models.

Performance Metrics

The lab provides objective measurements for model performance, including accuracy (pass rate on objective tests) and speed (tokens per second). These metrics are calculated based on runs conducted within the user's browser, with results remaining local to the machine. Charts display the latest suite performance per model.

Benchmark Customization and Certification

Users can write custom benchmarks using JavaScript within an editor provided by the platform. Each check runs against the model's decoded text. Additionally, the platform allows for the generation and download of verifiable performance certificates, detailing device hardware and sustained tokens per second, which can then be shared.