Oumi: Open Source LLM/VLM Training Platform

Oumi is a fully open-source platform designed to streamline the entire lifecycle of foundation models, from data preparation and training to evaluation and deployment.

Key features:

• Model Training - Train and fine-tune models ranging from 10M to 405B parameters using state-of-the-art techniques such as SFT, LoRA, and QLoRA.
• Multimodal Support - Work with both text and multimodal models, including Llama, DeepSeek, and Qwen.
• Data Synthesis - Synthesize and curate training data with built-in LLM judges.
• Efficient Deployment - Deploy models using popular inference engines like vLLM and SGLang.
• Comprehensive Evaluation - Evaluate models across standard benchmarks.
• Flexible Usage - Run on various platforms, from laptops to cloud services (AWS, Azure, GCP).

Oumi is suitable for researchers and developers looking for a comprehensive platform to train, evaluate, and deploy machine learning models efficiently.