█

AI/TLDR

Oumi

Open-source platform for training, fine-tuning, and deploying foundation models end-to-end

GitHub★ 9.4kWebsiteoumi.ai
Fine-Tuning FrameworksOpen source
Language
Python
License
Apache-2.0

Overview

Oumi is a fully open-source platform that handles the entire lifecycle of foundation models, from preparing data and training to evaluation and deployment. It works the same way whether you run it on a laptop, on a cluster, or in production.

You can train and fine-tune models from 10M up to 405B parameters using techniques like SFT, LoRA, QLoRA, and GRPO. Oumi supports both text and multimodal models such as Llama, DeepSeek, Qwen, and Phi, and it ships ready-to-use recipes so you do not have to write training loops or data pipelines by hand.

The same consistent API connects open models and commercial APIs (OpenAI, Anthropic, Vertex AI, Together, and more), and lets you deploy with popular inference engines like vLLM and SGLang. Jobs can run locally or on AWS, Azure, GCP, and Lambda.

What it does

  • Train and fine-tune models from 10M to 405B parameters using SFT, LoRA, QLoRA, GRPO, and other techniques
  • Work with both text and multimodal models including Llama, DeepSeek, Qwen, and Phi
  • Synthesize and curate training data with LLM judges
  • Deploy efficiently with inference engines like vLLM and SGLang
  • Evaluate models across standard benchmarks with one consistent API
  • Run anywhere, from laptops to clusters to clouds (AWS, Azure, GCP, Lambda)

Getting started

Install Oumi with pip (uv is recommended), then use the oumi CLI to train, evaluate, and run inference with one of the included recipes.

Install Oumi

Install the base package, or add the gpu extra for GPU support. uv is recommended, but plain pip works too.

bashbash
# Basic installation
uv pip install oumi

# With GPU support
uv pip install 'oumi[gpu]'

Train a model from a recipe

Point the oumi CLI at a recipe config to start training. The repo ships many ready-to-use recipes under configs/recipes.

bashbash
oumi train -c configs/recipes/smollm/sft/135m/quickstart_train.yaml

Evaluate and run inference

Use the same CLI to evaluate a model against benchmarks or run interactive inference.

bashbash
# Evaluation
oumi evaluate -c configs/recipes/smollm/evaluation/135m/quickstart_eval.yaml

# Inference
oumi infer -c configs/recipes/smollm/inference/135m_infer.yaml --interactive

Run jobs remotely on the cloud

Use oumi launch to run jobs on cloud platforms such as GCP, AWS, Azure, or Lambda.

bashbash
oumi launch up -c configs/recipes/smollm/sft/135m/quickstart_gcp_job.yaml --resources.cloud aws

Commands and code are distilled from the project's own documentation — always check the official repo for the latest.

When to use it

  • Fine-tune an open model such as Llama, Qwen, or DeepSeek for your own task using SFT, LoRA, or QLoRA recipes
  • Synthesize and curate training data with LLM judges before training
  • Evaluate a model across standard benchmarks with a single consistent API
  • Deploy a fine-tuned model with vLLM or SGLang and launch training jobs on AWS, Azure, GCP, or Lambda

How Oumi compares

Oumi alongside other open-source fine-tuning frameworks tools AI/TLDR tracks, ranked by GitHub stars.

ToolStarsWhat it does
Unsloth★ 76.4kA library that speeds up LoRA and QLoRA fine-tuning while cutting memory use, aimed at training models on a single GPU.
LLaMA-Factory★ 74.9kAn end-to-end training suite with a web UI that covers pre-training, supervised fine-tuning, and RLHF for hundreds of LLMs and multimodal models.
PEFT★ 21.7kHugging Face's library of parameter-efficient fine-tuning methods such as LoRA, DoRA, and prompt tuning that train small adapters instead of full models.
FinGPT★ 21.3kFinGPT is an open-source project of financial LLMs, fine-tuned with LoRA on news and tweet data for tasks like sentiment analysis, relation extraction, and stock-move forecasting.
ms-swift★ 15.7kModelScope's framework for fine-tuning and deploying 600+ LLMs and 300+ multimodal models, supporting PEFT and full-parameter SFT, DPO, and GRPO.
LitGPT★ 13.7kAn open-source toolkit from Lightning AI to pretrain, finetune, and serve 20+ large language models, each written from scratch for speed and full control.
Axolotl★ 12.5kA config-driven tool for fine-tuning and post-training open LLMs that supports SFT, LoRA/QLoRA, DPO, GRPO, and multi-GPU training across many model families.
Oumi★ 9.4kOpen-source platform for training, fine-tuning, and deploying foundation models end-to-end