15 Pull requests merged by 5 people

• [Evaluation] Updates in hallucination notebook

  #1552 merged Mar 19, 2025

• Remove UV install from notebooks as this breaks colab

  #1558 merged Mar 19, 2025

• Add letter counting GRPO example

  #1539 merged Mar 19, 2025

• Add OUMI_FORCE_EDITABLE_INSTALL env var to do editable Oumi install from source in job configs

  #1420 merged Mar 19, 2025

• Add code used for the evaluation demo.

  #1556 merged Mar 18, 2025

• Misc no-op code cleanups

  #1553 merged Mar 18, 2025

• Augmenting logging training/model statistics

  #1545 merged Mar 18, 2025

• [Evaluation] Modifying Alpaca Eval results format to be consistent with LM Harness

  #1551 merged Mar 17, 2025

• Define Conversation proto bufs

  #1550 merged Mar 17, 2025

• Render the oumi env command as a shell block in bug reports.

  #1548 merged Mar 17, 2025

• Update the CLI to replace all prints with Rich prints.

  #1547 merged Mar 17, 2025

• Update oumi evaluate to use rich formatting.

  #1544 merged Mar 16, 2025

• Update oumi launch to use Rich formatting

  #1543 merged Mar 16, 2025

• Update Oumi Env to use Rich formatting

  #1541 merged Mar 14, 2025

• Auto-populate and validate params specific to vision_language_sft collator in TrainingConfig

  #1537 merged Mar 13, 2025

4 Pull requests opened by 4 people

• Add KTO support for preference tuning

  #1538 opened Mar 13, 2025

• [DRAFT-Not-for-full-review] phi 4 multimodal

  #1555 opened Mar 18, 2025

• [Evaluations] Custom evals: Adding support for `eval_kwargs`

  #1557 opened Mar 19, 2025

• Support text truncation (`max_length`) for `vision_language_sft` collator

  #1559 opened Mar 19, 2025

2 Issues closed by 1 person

• [Bug] Runtime error in google colab

  #1554 closed Mar 19, 2025

• [Bug] Test

  #1549 closed Mar 17, 2025

3 Issues opened by 2 people

• [Feature] oumi evaluate should save the evaluation config in the output_dir

  #1546 opened Mar 17, 2025

• [Bug] Examples in a GRPO training batch are identical

  #1542 opened Mar 15, 2025

• [Bug] `training.include_performance_metrics` does not work for GRPO

  #1540 opened Mar 14, 2025

4 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• Added 3 Pixmo vision-language datasets

  #1523 commented on Mar 19, 2025 • 26 new comments

• Pillow 10.3.0 not supported for Python 3.13

  #1525 commented on Mar 19, 2025 • 0 new comments

• Add tokenized example debugging during training

  #1520 commented on Mar 15, 2025 • 0 new comments

• Add s1k dataset implementation

  #1527 commented on Mar 15, 2025 • 0 new comments