Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Shen, Jucheng, Sarkar, Gaurav, Ro, Yeonju, Sridhar, Sharath Nittur, Wang, Zhangyang, Akella, Aditya, Kundu, Souvik
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2512.07173
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866917414922354688
author Shen, Jucheng
Sarkar, Gaurav
Ro, Yeonju
Sridhar, Sharath Nittur
Wang, Zhangyang
Akella, Aditya
Kundu, Souvik
author_facet Shen, Jucheng
Sarkar, Gaurav
Ro, Yeonju
Sridhar, Sharath Nittur
Wang, Zhangyang
Akella, Aditya
Kundu, Souvik
contents We present CadLLM, a training-free method to accelerate the inference throughput of diffusion-based LLMs (dLLMs). We first investigate the dynamic nature of token unmasking confidence across blocks and steps. Based on this observation, we present a lightweight adaptive approach that controls the generation block size, step size, and threshold based on the average confidence of unmasked tokens. We further reduce softmax overhead by dynamically leveraging a subset of the vocabulary to regulate sampling breadth. CadLLM is a plug-and-play, model-agnostic method compatible with KV-cache-based dLLMs. Extensive experiments on four popular tasks demonstrate that CadLLM yields up to 1.1-2.28x throughput improvement over the state-of-the-art baseline with competitive accuracy.
format Preprint
id arxiv_https___arxiv_org_abs_2512_07173
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Improving the Throughput of Diffusion-based Large Language Models via a Training-Free Confidence-Aware Calibration
Shen, Jucheng
Sarkar, Gaurav
Ro, Yeonju
Sridhar, Sharath Nittur
Wang, Zhangyang
Akella, Aditya
Kundu, Souvik
Machine Learning
We present CadLLM, a training-free method to accelerate the inference throughput of diffusion-based LLMs (dLLMs). We first investigate the dynamic nature of token unmasking confidence across blocks and steps. Based on this observation, we present a lightweight adaptive approach that controls the generation block size, step size, and threshold based on the average confidence of unmasked tokens. We further reduce softmax overhead by dynamically leveraging a subset of the vocabulary to regulate sampling breadth. CadLLM is a plug-and-play, model-agnostic method compatible with KV-cache-based dLLMs. Extensive experiments on four popular tasks demonstrate that CadLLM yields up to 1.1-2.28x throughput improvement over the state-of-the-art baseline with competitive accuracy.
title Improving the Throughput of Diffusion-based Large Language Models via a Training-Free Confidence-Aware Calibration
topic Machine Learning
url https://arxiv.org/abs/2512.07173