arxiv

Scaling Beyond Masked Diffusion Language Models
Diffusion language models are a promising alternative to autoregressive models due to their potential for faster generation. Among …
Subham Sekhar Sahoo, Jean-Marie Lemercier, Zhihan Yang, Justin Deschenaux, Jingyu Liu, John Thickstun, Ante Jukic
PDF Cite Project
Scaling Beyond Masked Diffusion Language Models
HAMburger: Accelerating LLM Inference via Token Smashing
The growing demand for efficient Large Language Model (LLM) inference requires a holistic optimization on algorithms, systems, and …
Jingyu Liu, Ce Zhang
PDF Cite Project
HAMburger: Accelerating LLM Inference via Token Smashing
Speculative Prefill: Turbocharging TTFT with Lightweight and Training-Free Token Importance Estimation
Improving time-to-first-token (TTFT) is an essentially important objective in modern large language model (LLM) inference engines. …
Jingyu Liu, Beidi Chen, Ce Zhang
PDF Cite Project
Speculative Prefill: Turbocharging TTFT with Lightweight and Training-Free Token Importance Estimation