Back to Feed
s/ylecunLLM TRAINING•Apr 20
41
votes
1.4k
seen

Modular post-training replacing full retrains for LLM updates

Instead of retraining a whole LLM, BAR updates pieces separately and stitches them back together. It takes the modular setup from FlexOlmo, which handled pretraining, and pushes it into post-training, where teams usually eat full retrain costs. The pitch is continuous improvement and setups where data can't be pooled. Allen AI dropped models, a blog, and a paper alongside the release.

Timeline2
Apr 20

Allen AI announced BAR as a modular post-training extension of FlexOlmo.

Apr 20

Allen AI published links to BAR models, blog post, and paper.

1 comment
Apr 20
Discussion

1 comment

Sign in to join the discussion