oumi Releases

Focuses on adaptive inference infrastructure, synthesis improvements, and fixes for the Judge API and configuration errors.

Jul 11, 2025

Adds GRPO training support for trl and verl, releases the HallOumi and CoALM models, supports new model architectures (Llama 4, Qwen3, Falcon H1), and adds Slurm/Frontier cluster support.

Jun 23, 2025

Adds Falcon-E integration, Molmo model configs, vision processing improvements, and support for data synthesis configurations.

Jun 10, 2025

Integrates verl PPO and GRPO trainers, adds Qwen3 model support, introduces the Geometry3K dataset, and adds Phi4 reasoning model configs.

May 29, 2025

Updates core dependencies (Transformers 4.51.0, TRL 0.16), adds Pixmo vision-language datasets, and improves remote inference error handling.

Apr 16, 2025

Adds Llama 4 support, LoRA training for Phi3/Phi4/Qwen2.5-VL, introduces MLflow integration, and adds HallOumi inference notebooks.

Apr 6, 2025

Fixes compatibility issues with Python 3.13 and resolves a bug regarding multiple GPUs attempting to save results in the same folder.

Mar 25, 2025

This release introduces QLoRA and QwQ configurations, updates the TRL library to 0.15, and adds support for Phi-4 multimodal training and inference.

Mar 24, 2025

This release introduces the GRPO trainer, modularizes the evaluation system, and adds support for the `oumi://` prefix in CLI commands.

Mar 10, 2025

This release focuses on expanding the library's icon set with new AWS services and adds support for new providers and tools. It also includes dependency updates and bug fixes.

Mar 10, 2025

Kubero adds several new application templates and enhances the addon management interface to include versioning and better visibility of configured addons.

Feb 26, 2025

This release adds support for Claude Sonnet 3.7, migrates the LM Harness integration, and includes a notebook for fine-tuning MiniMath-R1-1.5B.

Feb 25, 2025

This release enhances VLM training by supporting multi-image inputs and fixes the launcher to work on RunPod and Lambda cloud providers.

Feb 22, 2025

This release adds support for several new datasets (Docmatix, the_cauldron, Tulu3), integrates SambaNova for inference, and adds SGLang multi-image support.

Feb 20, 2025

This release adds DeepSeek R1 configurations, updates VLM notebooks, and fixes the Qwen2-VL evaluation job.

Feb 3, 2025

This release adds VLM/LoRA support, improves GGUF support in VLLM, and adds integration tests for evaluation functions.

Jan 28, 2025

This release focuses on bug fixes, including improvements to GGUF support and pyright compatibility, along with various documentation updates.

Jan 27, 2025

The update introduces new network and AI-related icons while updating the minimum required Python version and adding new tools like SQL and n8n.

Nov 27, 2024

This maintenance release upgrades the project to support Python 3.12 and 3.13, ensuring compatibility while fixing tests and maintaining type hint backward compatibility.

Nov 7, 2024

Feroxbuster gains new command-line options to support more complex scanning scenarios, including reading raw HTTP requests and forcing recursion into directory listings.

Sep 15, 2024