Running Modern LLMs in the Browser with Three.js and WebGPU
Three-LLM runs GPT-2 and other models in the browser using Three.js and WebGPU.
Three-LLM enables the local execution of GPT-2, SmolLM2, Qwen, and Phi models in the browser. This project was initiated to explore the compute capabilities of Three.js and WebGPU. The models range from a 3-million parameter story generator to an 800-million parameter chat model. Three.js extends beyond rendering, providing essential compute capabilities for language models.
This synthesis was produced from its source by AI; there is no human editor or manual review step. How we work