A developer has created SLMGen, an open-source tool that automates the setup and fine-tuning of Small Language Models like Llama 3.2 and Phi-4. The project tackles the common friction of manual Colab setup, data formatting, and model selection by providing a unified Python and Next.js application. Its core is a custom recommendation engine that scores models based on task, deployment, and data features.

The tool's recommendation engine uses a 100-point scoring system to match datasets with the optimal model, considering factors like multilingual support for Qwen 2.5 or edge deployment penalties for larger models. It also performs dataset intelligence scans for quality and hallucination risk. A key feature is generating self-contained Colab notebooks that embed data as base64, eliminating manual uploads and path errors.

SLMGen integrates with Unsloth for faster fine-tuning and supports 11 top-tier SLMs, including Gemma 2 and Mistral 7B. By automating the entire pipeline from data analysis to notebook generation, it aims to lower the barrier for developers to experiment with custom model training. The project is fully open source, with a live demo available for testing.