Starstruk Logo MarkStarstruk

v0.1.13

oumi-ai/oumiv0.1.13May 29, 2025by nikg7

AI Summary

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

Key Highlights

  • Verl PPO and GRPO trainer integration.
  • Qwen3 model support (32B, 30B A3B).
  • Geometry3K VLM dataset added.
  • Phi4 reasoning models added.

New Features

  • Verl PPO trainer implementation.
  • Verl GRPO configs and Countdown configs.
  • Qwen3 training and inference configs.
  • Geometry3K dataset integration.
  • Countdown dataset and reward function.
  • Phi4 reasoning configs.
  • Lambda inference engine support.
  • Frontier launcher script.

Full Release Notes

## What's Changed
* Update dev_setup.md by @wizeng23 in https://github.com/oumi-ai/oumi/pull/1641
* [tiny] Remove vllm install commands by @wizeng23 in https://github.com/oumi-ai/oumi/pull/1643
* Support for custom `processor args`: misc improvements by @nikg4 in https://github.com/oumi-ai/oumi/pull/1642
* Add Countdown dataset and reward function by @wizeng23 in https://github.com/oumi-ai/oumi/pull/1645
* Adding LoRA train config for Qwen-VL 2.0 by @optas in https://github.com/oumi-ai/oumi/pull/1637
* [Evaluation] Convenience function for standard config retrieval by @kaisopos in https://github.com/oumi-ai/oumi/pull/1644
* Add demo script by @oelachqar in https://github.com/oumi-ai/oumi/pull/1647
* [bug] fix build errors by @oelachqar in https://github.com/oumi-ai/oumi/pull/1649
* Adding LoRA train config for SmolVLM by @optas in https://github.com/oumi-ai/oumi/pull/1639
* [tiny] Update cli help shorthand by @oelachqar in https://github.com/oumi-ai/oumi/pull/1648
* Oelachqar/update hooks by @oelachqar in https://github.com/oumi-ai/oumi/pull/1650
* Add verl PPO trainer by @wizeng23 in https://github.com/oumi-ai/oumi/pull/1646
* Fix a missing dependency in the verl trainer. by @taenin in https://github.com/oumi-ai/oumi/pull/1651
* Integrate verl GRPO trainer into train script by @wizeng23 in https://github.com/oumi-ai/oumi/pull/1652
* Update e2e tests to run on lambda by @wizeng23 in https://github.com/oumi-ai/oumi/pull/1653
* Add Qwen3 32B configs by @wizeng23 in https://github.com/oumi-ai/oumi/pull/1661
* Add Qwen3 30B A3B configs by @wizeng23 in https://github.com/oumi-ai/oumi/pull/1665
* [verl] Populate verl config from Oumi config by @wizeng23 in https://github.com/oumi-ai/oumi/pull/1659
* Provide option to configure `label_ignore_index` in training config by @nikg4 in https://github.com/oumi-ai/oumi/pull/1666
* [Documentation] Custom Evaluations (PR 1-of-2) by @kaisopos in https://github.com/oumi-ai/oumi/pull/1664
* InterVL-3.0 SFT with *limited* training capabilities by @optas in https://github.com/oumi-ai/oumi/pull/1663
* Add verl GRPO Countdown configs by @wizeng23 in https://github.com/oumi-ai/oumi/pull/1668
* Set explicit permissions for our test workflows. by @taenin in https://github.com/oumi-ai/oumi/pull/1670
* Add support for repetition_penalty in GrpoParams by @REDDITARUN in https://github.com/oumi-ai/oumi/pull/1654
* Fix broken tests due to precommit violations by @taenin in https://github.com/oumi-ai/oumi/pull/1671
* [Documentation] Custom Evaluations (PR 2-of-2) by @kaisopos in https://github.com/oumi-ai/oumi/pull/1669
* Migrate to `logger.warning` usage by @emmanuel-ferdman in https://github.com/oumi-ai/oumi/pull/1673
* Update the Oumi launcher and e2e tests to support runpod. by @taenin in https://github.com/oumi-ai/oumi/pull/1672
* Switch back to using GCP for e2e tests. by @taenin in https://github.com/oumi-ai/oumi/pull/1675
* Mark an e2e test as is_lora by @taenin in https://github.com/oumi-ai/oumi/pull/1676
* Add Phi4 reasoning plus configs by @wizeng23 in https://github.com/oumi-ai/oumi/pull/1674
* Fix a test breakage caused by a new Click version (8.2.0) by @taenin in https://github.com/oumi-ai/oumi/pull/1679
* chore: edited the link to the stars badge by @Radovenchyk in https://github.com/oumi-ai/oumi/pull/1681
* Update verl GRPO countdown configs by @wizeng23 in https://github.com/oumi-ai/oumi/pull/1682
* [very nit] center oumi logo in the cli by @oelachqar in https://github.com/oumi-ai/oumi/pull/1683
* [tiny] Update training environments doc by @wizeng23 in https://github.com/oumi-ai/oumi/pull/1686
* Add Geometry3K VLM dataset by @nikg4 in https://github.com/oumi-ai/oumi/pull/1687
* Add `torchao` version to `pyproject.toml` by @nikg4 in https://github.com/oumi-ai/oumi/pull/1688
* [Feature] Save evaluation config as YAML in output_dir #1546 by @asish-kun in https://github.com/oumi-ai/oumi/pull/1680
* Create a script to calculate memory used during training by @wizeng23 in https://github.com/oumi-ai/oumi/pull/1441
* Support VLM-s with VERL_GRPO trainer by @nikg4 in https://github.com/oumi-ai/oumi/pull/1689
* docs: Add GRPO/verl documentation by @wizeng23 in https://github.com/oumi-ai/oumi/pull/1690
* Update GRPO letter counting reward function and hparams for stability by @jgreer013 in https://github.com/oumi-ai/oumi/pull/1692
* [GRPO] Update letter counting notebook by @wizeng23 in https://github.com/oumi-ai/oumi/pull/1694
* Add Lambda Inference Engine by @oelachqar in https://github.com/oumi-ai/oumi/pull/1695
* Basic shell script for launching jobs on OLCF Frontier HPC cluster by @nikg4 in https://github.com/oumi-ai/oumi/pull/1691
* Add CoALM dataset class by @oelachqar in https://github.com/oumi-ai/oumi/pull/1696
* Added exponential backoff and content-type error handling in remote inference engine by @abhiramvad in https://github.com/oumi-ai/oumi/pull/1685
* Make SFT datasets usable with GRPO_TRL trainer by @nikg4 in https://github.com/oumi-ai/oumi/pull/1698
* Implement Falcon H1 by @dhiaEddineRhaiem in https://github.com/oumi-ai/oumi/pull/1699
* [tiny] Remove deprecated `use_async_dataset` from configs by @wizeng23 in https://github.com/oumi-ai/oumi/pull/1701
* Add sample inference configs for `HuggingFaceTB/SmolVLM-Instruct` by @nikg4 in https://github.com/oumi-ai/oumi/pull/1703

## New Contributors
* @REDDITARUN made their first contribution in https://github.com/oumi-ai/oumi/pull/1654
* @emmanuel-ferdman made their first contribution in https://github.com/oumi-ai/oumi/pull/1673
* @Radovenchyk made their first contribution in https://github.com/oumi-ai/oumi/pull/1681
* @asish-kun made their first contribution in https://github.com/oumi-ai/oumi/pull/1680
* @abhiramvad made their first contribution in https://github.com/oumi-ai/oumi/pull/1685
* @dhiaEddineRhaiem made their first contribution in https://github.com/oumi-ai/oumi/pull/1699

**Full Changelog**: https://github.com/oumi-ai/oumi/compare/v0.1.12...v0.1.13