IBM Research trains a 20B open model to build decks in a house style

IBM Research's Palette fine-tunes a 20B open-weight model with a LoRA adapter to make slide decks that follow a set design language, here IBM's own. One role plans the deck as an editable JSON brief and another writes each slide's code, then a checker renders every slide and keeps a repair only if it actually fixes overlaps or text running off the slide. On the PresentBench benchmark it ranked first on two of three domains and second overall, 1.1 points behind NotebookLM (59.7 vs 60.8). It was trained on about 1,000 synthetic decks; this is a research write-up, and the post doesn't say whether the model is released.