Nathan Pemberton

Mude Hui, Xin Huang, Jaime Campos Salas et al.

Reasoning LLMs produce longer outputs, requiring speculative decoding drafters trained on extended sequences. Parallel drafting - predicting multiple tokens per forward pass - offers latency benefits over sequential generation, but training complexity scales quadratically with the product of sequence length and parallel positions, rendering long-context training impractical. We present P(arallel)-EAGLE, which transforms EAGLE from autoregressive to parallel multi-token prediction via a learnable shared hidden state. To scale training to long contexts, we develop a framework featuring attention mask pre-computation and sequence partitioning techniques, enabling gradient accumulation within individual sequences for parallel-prediction training. We implement P-EAGLE in vLLM and demonstrate speedups of 1.10-1.36x over autoregressive EAGLE-3 across GPT-OSS 120B, 20B, and Qwen3-Coder 30B.

David Culler, Prabal Dutta, Gabe Fierro et al.

Governments around the world have become increasingly frustrated with tech giants dictating public health policy. The software created by Apple and Google enables individuals to track their own potential exposure through collated exposure notifications. However, the same software prohibits location tracking, denying key information needed by public health officials for robust contract tracing. This information is needed to treat and isolate COVID-19 positive people, identify transmission hotspots, and protect against continued spread of infection. In this article, we present two simple ideas: the lighthouse and the covid-commons that address the needs of public health authorities while preserving the privacy-sensitive goals of the Apple and google exposure notification protocols.