Encoder-Decoder Diffusion Language Models for Efficient Training and Inference
This work addresses efficiency bottlenecks in diffusion language models for NLP applications, offering faster training and inference, though it is incremental as it builds on existing diffusion and block diffusion methods.
The paper tackles the high computational cost of discrete diffusion models in language generation by proposing an encoder-decoder architecture that separates clean token representation and denoising tasks, achieving superior trade-offs between generation quality and inference throughput on summarization, translation, and mathematical reasoning tasks.
Discrete diffusion models enable parallel token sampling for faster inference than autoregressive approaches. However, prior diffusion models use a decoder-only architecture, which requires sampling algorithms that invoke the full network at every denoising step and incur high computational cost. Our key insight is that discrete diffusion models perform two types of computation: 1) representing clean tokens and 2) denoising corrupted tokens, which enables us to use separate modules for each task. We propose an encoder-decoder architecture to accelerate discrete diffusion inference, which relies on an encoder to represent clean tokens and a lightweight decoder to iteratively refine a noised sequence. We also show that this architecture enables faster training of block diffusion models, which partition sequences into blocks for better quality and are commonly used in diffusion language model inference. We introduce a framework for Efficient Encoder-Decoder Diffusion (E2D2), consisting of an architecture with specialized training and sampling algorithms, and we show that E2D2 achieves superior trade-offs between generation quality and inference throughput on summarization, translation, and mathematical reasoning tasks. We provide the code, model weights, and blog post on the project page: https://m-arriola.com/e2d2