Credits & Acknowledgments

What it does:

WebLLM is the entire foundation of this project. It enables running large language models completely in your browser using WebGPU for GPU acceleration. Without WebLLM, none of the benchmarking, model loading, or inference would be possible.

How we use it:

• •Loading quantized LLM models directly in the browser
• •Running GPU-accelerated inference using WebGPU
• •Measuring tokens per second, load times, and performance metrics
• •Detecting GPU capabilities and system specifications

What it does:

Hugging Face is the platform hosting and distributing all the AI models used in our benchmarks. They provide the infrastructure to access pre-quantized models optimized for browser execution. Without their model repository and CDN, we couldn't load models efficiently.

How we use it:

• •Downloading MLC-quantized model files on-demand
• •Accessing models from Meta, Microsoft, Alibaba, and others
• •Leveraging their global CDN for fast model delivery

Models we benchmark: