PG-19

The PG-19 dataset is a benchmark used to evaluate whole-book language modeling tasks.

Papers using PG-19 (3)

Dustin: Draft-Augmented Sparse Verification for Efficient Long-Context Generation with Speculative Decoding2026Locas: Your Models are Principled Initializers of Locally-Supported Parametric Memories2026MTraining: Distributed Dynamic Sparse Attention for Efficient Ultra-Long Context Training2025