Starstruk Logo MarkStarstruk

v0.1.14

oumi-ai/oumiv0.1.14Jun 10, 2025by wizeng23

AI Summary

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

Key Highlights

  • Falcon-E integration and configs.
  • Molmo model support (configs and GRPO).
  • Vision processing improvements (individual image processing).
  • Data synthesis config and unit tests.

New Features

  • Falcon-E configs and LoRA examples.
  • Molmo configs (O variant, GRPO).
  • Vision dataset generation improvements.
  • Data synthesis configuration support.
  • Remote inference exception handling and fast-failing.
  • DPO + QLoRA example for Falcon-H1.

Full Release Notes

## What's Changed
* Record latency histograms in base inference engine by @nikg4 in https://github.com/oumi-ai/oumi/pull/1702
* Feat: add falcon-e integration by @younesbelkada in https://github.com/oumi-ai/oumi/pull/1705
* [tiny] Minor update to fix the failing pre-commit checks by @oelachqar in https://github.com/oumi-ai/oumi/pull/1707
* Add collator kwargs field to DataParams by @oelachqar in https://github.com/oumi-ai/oumi/pull/1708
* [vision] Add option to process images individually by @oelachqar in https://github.com/oumi-ai/oumi/pull/1706
* Update dev_setup.md to correct the order of steps by @ryan-arman in https://github.com/oumi-ai/oumi/pull/1709
* Add configs for molmo support by @oelachqar in https://github.com/oumi-ai/oumi/pull/1710
* [tiny] fix pre-commits checks on a fresh install by @oelachqar in https://github.com/oumi-ai/oumi/pull/1711
* Add config for the Molmo O variant by @oelachqar in https://github.com/oumi-ai/oumi/pull/1712
* Add experimental molmo grpo config and train aliases by @oelachqar in https://github.com/oumi-ai/oumi/pull/1713
* Update installation.md to fix subversion handling by adding required … by @ryan-arman in https://github.com/oumi-ai/oumi/pull/1715
* Frontier: Fix -n param in launcher script by @nikg4 in https://github.com/oumi-ai/oumi/pull/1720
* Fix Falcon H1 dependency setup by @wizeng23 in https://github.com/oumi-ai/oumi/pull/1723
* letter count notebook improvements by @penfever in https://github.com/oumi-ai/oumi/pull/1697
* [vision] Update vision feature generator to support training on completions only by @oelachqar in https://github.com/oumi-ai/oumi/pull/1722
* [tiny] fix bug with vl collator  by @oelachqar in https://github.com/oumi-ai/oumi/pull/1725
* Add data synthesis config, params, and unit tests by @jgreer013 in https://github.com/oumi-ai/oumi/pull/1700
* Add support for additional exception types for remote inference engine, as well as fast failing for non-retryable status codes. by @jgreer013 in https://github.com/oumi-ai/oumi/pull/1704
* Adds DPO + QLoRA example for Falcon-H1 by @stefanwebb in https://github.com/oumi-ai/oumi/pull/1719
* Update inference to always write intermediate results to file. by @jgreer013 in https://github.com/oumi-ai/oumi/pull/1724
* Added doc for new QLoRA param by @stefanwebb in https://github.com/oumi-ai/oumi/pull/1727
* Readme for Falcon-E and note on extra dependencies required by @stefanwebb in https://github.com/oumi-ai/oumi/pull/1729
* Add generic vision dataset by @oelachqar in https://github.com/oumi-ai/oumi/pull/1726
* [tiny][bug] make git cmd optional by @oelachqar in https://github.com/oumi-ai/oumi/pull/1730
* [tiny][bug] Add missing molmo feature by @oelachqar in https://github.com/oumi-ai/oumi/pull/1731
* [tiny] Update phi3-vision configs to use oumi trainer by @oelachqar in https://github.com/oumi-ai/oumi/pull/1733
* Minor bugfixes for 2 clouds in launcher code by @nikg4 in https://github.com/oumi-ai/oumi/pull/1728
* Update dev_setup.md to add additional instructions by @ryan-arman in https://github.com/oumi-ai/oumi/pull/1736
* Update trl to 0.18 by @wizeng23 in https://github.com/oumi-ai/oumi/pull/1693
* Update Verl trainer to export models in HF format by @nikg4 in https://github.com/oumi-ai/oumi/pull/1714
* Add lmms-lab/multimodal-open-r1-8k-verified dataset by @oelachqar in https://github.com/oumi-ai/oumi/pull/1732
* Add placeholders for DCVLR by @oelachqar in https://github.com/oumi-ai/oumi/pull/1738
* add debug logging capabilities to collators by @aniruddh-alt in https://github.com/oumi-ai/oumi/pull/1678
* [bug] update trainer to save processor when training with fsdp by @oelachqar in https://github.com/oumi-ai/oumi/pull/1742
* Add model revision param by @oelachqar in https://github.com/oumi-ai/oumi/pull/1740
* Add ability to customize HF model config via model.model_kwargs by @oelachqar in https://github.com/oumi-ai/oumi/pull/1741
* Add docker release workflow by @oelachqar in https://github.com/oumi-ai/oumi/pull/1743
* [bug] fix rank/local rank parsing for docker env by @oelachqar in https://github.com/oumi-ai/oumi/pull/1747
* deps: Update vLLM to 0.8.3 by @wizeng23 in https://github.com/oumi-ai/oumi/pull/1739
* [docs] update dcvlr readme by @oelachqar in https://github.com/oumi-ai/oumi/pull/1748
* Dcvlr by @penfever in https://github.com/oumi-ai/oumi/pull/1750

## New Contributors
* @younesbelkada made their first contribution in https://github.com/oumi-ai/oumi/pull/1705
* @ryan-arman made their first contribution in https://github.com/oumi-ai/oumi/pull/1709
* @stefanwebb made their first contribution in https://github.com/oumi-ai/oumi/pull/1719
* @aniruddh-alt made their first contribution in https://github.com/oumi-ai/oumi/pull/1678

**Full Changelog**: https://github.com/oumi-ai/oumi/compare/v0.1.13...v0.1.14