A demo presentation with math, code, and figures
Nathan Lambert
2026-02-22
The loss function for RLHF with a KL penalty:
Inline math works too: \nabla_\theta J(\theta) = \mathbb{E}_{\pi_\theta}[R \cdot \nabla_\theta \log \pi_\theta(a|s)]
from colloquium import Deck
deck = Deck(title="My Talk", author="Researcher")
deck.add_title_slide(subtitle="A research presentation")
deck.add_slide(
title="Key Results",
content="Our method achieves **state-of-the-art** performance.",
)
deck.build("output/")
This slide uses a two-column layout for side-by-side content.
Column 1 can have text and bullet points:
Column 2 can have different content:
| Model | Accuracy | F1 Score | Training Time | |-------|----------|----------|---------------| | Baseline | 82.1% | 79.3% | 2h | | Ours (small) | 91.4% | 89.7% | 4h | | Ours (large) | 95.2% | 93.8% | 12h |
"The results demonstrate significant improvements across all metrics."
Questions?
github.com/interconnects-ai/colloquium