WebLlama provides training scripts, optimized configs, and instructions for fine-tuning Llama models on web navigation tasks. The first model, Llama-3-8B-Web, is finetuned on over 24K instances of web interactions including clicks, text input, submission, and dialogue acts. The project includes evaluation through the WebLINX 1.0 benchmark for human-centric web browsing and integrates with deployment platforms like Playwright, Selenium, and BrowserGym.