Oumi: open-source platform for foundation models

Core facts

Read by a script from the official API, not rewritten by a model.

Repository: https://github.com/oumi-ai/oumi
License: Apache-2.0
Languages: Python
Maintainer: oumi-ai
Homepage: https://oumi.ai
Verified: 2026-11-01

What it is

Oumi is a fully open-source platform that streamlines the entire lifecycle of foundation models, from data preparation and training to evaluation and deployment. The project describes Oumi as a community-first effort. The project is licensed under the Apache License 2.0.

Main features

Oumi works with both text and multimodal models, including Llama, Gemma, Qwen, DeepSeek, GLM, Phi, and others.
Oumi synthesizes and curates training data, including multi-turn tool-use conversations, with LLM judges.
Oumi can deploy models to dedicated endpoints with the deploy command.
Oumi evaluates models across standard benchmarks and with multi-criteria rubric judges.
Oumi provides executable tool environments: database, HTTP endpoint, lookup, and simulated.

Getting started

The launch command runs jobs remotely on cloud platforms (AWS, Azure, GCP, Lambda, etc.).
The installation methods list using pip, marked as recommended.
The installation methods also list using Docker.
There is also a quick install script labeled experimental.

From the README · bash

# Basic installation
uv pip install oumi

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

# Latest development version
uv pip install git+https://github.com/oumi-ai/oumi.git

From the README · bash

# Pull the latest image
docker pull ghcr.io/oumi-ai/oumi:latest

# Run oumi commands
docker run --gpus all -it ghcr.io/oumi-ai/oumi:latest oumi --help

# Train with a mounted config
docker run --gpus all -v $(pwd):/workspace -it ghcr.io/oumi-ai/oumi:latest \
    oumi train --config /workspace/my_config.yaml

From the README · bash

curl -LsSf https://oumi.ai/install.sh | bash

From the README · shell

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

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

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

From the README · shell

# GCP
oumi launch up -c configs/recipes/smollm/sft/135m/quickstart_gcp_job.yaml

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

# Azure
oumi launch up -c configs/recipes/smollm/sft/135m/quickstart_gcp_job.yaml --resources.cloud azure

# Lambda
oumi launch up -c configs/recipes/smollm/sft/135m/quickstart_gcp_job.yaml --resources.cloud lambda

Who it is for

Oumi is in beta and under active development. The example configurations in the README are not an exhaustive list of what is supported. The README defines open models as models with fully open weights, training code, and data, and a permissive license.

Oumi runs from laptops to clusters to clouds (AWS, Azure, GCP, Lambda, Modal, Slurm, and more). Oumi integrates with open models and commercial APIs, including OpenAI, Anthropic, Vertex AI, Together, Fireworks, and Parasail. The oumi-mcp MCP server drives Oumi from Claude or Cursor.

The project says its Zero Boilerplate feature means no need to write training loops or data pipelines. Oumi has native support for distributed training techniques (FSDP, DeepSpeed, DDP). The project says Oumi is built and validated by teams training models at scale.

The news lists that GRPO reinforcement learning was extended to support tool use. The news lists that support for the Gemma 4 model family was added. The news lists a new deploy command for deploying Oumi models to dedicated inference endpoints on Fireworks and Parasail.

The news lists that support for the Qwen3.5 model family was added. The news lists training and inference support for OpenAI's gpt-oss-20b and gpt-oss-120b. The news lists that support for training and inference with Llama 4 models was added.

The news lists added support for tools, environments (simulated, lookup, database), and agentic data synthesis.

Related tools listed on this site.