🤗 Hugging Face - a treasure trove of models #MLSTART

If Groq and OpenRouter are platforms where you can test models, then Hugging Face is a whole ecosystem. It has everything: models, datasets, pipelines, demos, and even training tutorials.

💫 The coolest thing is that Hugging Face can be considered as GitHub for ML. Every developer can upload their model there, and you can take it, test it, or fine-tune it for your task.

For example, the folks at Vikhrmodels have several excellent Russian-language models. I personally used Vikhr-Nemo-12B-Instruct-R-21-09-24 - it handles Russian tasks quite confidently. However, it might be a bit heavy for an average PC, so you may need quantization.

Among other open-source options, Qwen performs well - decent generation quality, with "light" versions available, so you can run it on your computer.

By the way, for local LLM deployment with a chat interface, there are many services. My favorite is LM Studio.

⚡️ If you want to experiment with local models or try your hand at fine-tuning - welcome to Hugging Face!

Back to table of contents

👩‍💻 Data Flow