IBM fine-tuned 20B model Palette generates slides
The story so far
IBM's research team fine-tuned a 20B open-weight model Palette to generate slides according to IBM Design Language. The approach involves two LoRA roles: the designer role writes JSON briefings, and the encoder role generates pptxgenjs code page by page. On PresentBench, Palette ranked first in two of the three measured domains; it ranked second overall, scoring 59.7, 1.1 points lower than NotebookLM's 60.8, outperforming other models. The results are for a single 20B open-weight model. In the test, PresentBench only provides source documents, not plans, so another cutting-edge model drafts each page plan based on the document, and Palette then generates the slides accordingly.
AI-written from the reports · Updated 14 days ago
Coverage timeline
Follow the coverage to see every side of the story.
- Hugging Face BlogFeaturedPalette: teaching a small model to build decks in your design language
IBM's research team fine-tunes a 20B open-source weight model, Palette, using a two-role LoRA scheme where designers write JSON briefs and encoders generate pptxgenjs code page by page to create presentations according to IBM Design Language.