Palette: 20B open-weight model with LoRA adapter builds slide decks in a company's design language
Overview
Hugging Face describes Palette, a 20B open-weight model fine-tuned with a LoRA adapter to generate slide decks in a company's design language.
The work is split between a designer role that writes a JSON brief and a coder role that writes pptxgenjs code for each slide. The team says Palette ranks first on two of three domains on PresentBench, and that per-slide coding runs about 6 to 7 times faster through parallel calls. The report is a first-party Hugging Face blog post; the performance claims are the company's own and have not been independently verified.
The report credits the Palette work to Hugging Face, while the attached fact record names IBM Research as the subject. The record's conditions and evidence fields are empty, so the attribution to IBM Research is not supported by the report itself.
Written by AI from the articles below · updated Nov 3, 12:25 PM ET
Check the sources:
Article timeline
The articles in this story. Times are ET.
- Hugging Face BlogOfficialHugging Face's Palette fine-tunes a 20B model to build decks in your design language
AIHugging Face trains a 20B open-weight model with a LoRA adapter to generate slide decks in a company's design language, splitting work between a designer role that writes a JSON brief and a coder role that writes pptxgenjs code per slide. The team says the model ranks first on two of three domains on PresentBench, and that per-slide coding runs about 6–7 times faster through parallel calls.
Heat Trend
Not enough continuous observations to show a trend yet.