HomeToolsMCPHow It WorksStoriesPhilosophyCommunityArchitectureStar on GitHub
All Tools
O
Fine-tuningFreeOpen Source

OUMI

Everything you need to build state-of-the-art foundation models, end-to-end

Apache-2.0

ABOUT

Fine-tuning and deploying foundation models involves a fragmented toolchain — separate scripts for data prep, training, evaluation, and inference across different hardware configurations. Oumi unifies this lifecycle by providing one configuration-driven platform that handles data synthesis, LLM-as-a-judge curation, distributed training (SFT/DPO/GRPO/LoRA), standard benchmarks, and vLLM/SGLang-based deployment — from laptop to multi-node clusters.

INSTALL
pip install oumi

INTEGRATION GUIDE

1. Full fine-tuning and LoRA/QLoRA tuning of open-source LLMs like Llama 4 and DeepSeek 2. GRPO/RLHF fine-tuning for improving reasoning model performance 3. Synthesize training data and curate datasets with LLM-as-a-judge evaluation 4. Evaluate models across standard LLM and VLM benchmarks in a consistent pipeline 5. Deploy fine-tuned models with optimized inference engines (vLLM, SGLang) to production

TAGS

fine-tuningsftdpoloraqloragrpotrainingllmevaluationinferencevlmdeepseekllamaqwen