Palette Trains a 20B Model to Build Slides in IBM’s Design Language
Palette: teaching a small model to build decks in your design language
- Who
- Hugging Face and IBM Research
- Quote
This post covers how we synthesized the data, trained the model and built a harness around it.
What happened
IBM Research introduces Palette, a fine-tuned 20B open-weight model that builds slide decks using IBM Design Language. It divides the work between a designer that creates a structured brief and a coder that generates slide code, with a harness that checks and repairs rendering problems. On PresentBench, Palette ranks first in Economics and Education, third in Talk, and second on the three-domain mean.
Why it matters
Palette separates deck planning from slide geometry and uses a render-and-repair harness to check generated slides. The reported benchmark results show where this approach performs well across three domains.
Read next
See the full storyIBM Research Introduces Palette, a 20B Model for IBM-Style Slide DecksThe story so far
IBM Research introduced Palette, a fine-tuned 20B open-weight model and harness that build slide decks in IBM Design Language. Palette splits the work between a designer that creates a structured brief and a coder that generates slide code; the harness checks and repairs rendering problems. On PresentBench, Palette ranked first in Economics and Education, third in Talk, and second on the mean across those three domains.
Keep up with AI, one email a day
Model releases, price changes and compute developments, with original sources. Free, confirmed by email, and easy to unsubscribe. Read a past issue · 2026-11-03