Large Language Diffusion Models
This work is significant for the natural language processing community as it provides an alternative to traditional autoregressive models, which is an incremental yet important step for researchers and developers working on large language models.
The authors introduced LLaDA, a diffusion model that challenges the notion that large language models rely on autoregressive models, and demonstrated its strong scalability and competitive performance with state-of-the-art models like LLaMA3 8B. LLaDA 8B was competitive in in-context learning and instruction-following abilities, and even surpassed GPT-4o in a reversal poem completion task.
The capabilities of large language models (LLMs) are widely regarded as relying on autoregressive models (ARMs). We challenge this notion by introducing LLaDA, a diffusion model trained from scratch under the pre-training and supervised fine-tuning (SFT) paradigm. LLaDA employs a forward data masking process and a reverse generation process, parameterized by a Transformer to predict masked tokens. It provides a principled generative approach for probabilistic inference by optimizing a likelihood lower bound. Across extensive benchmarks on general tasks, math, code, and so on, LLaDA demonstrates strong scalability and performs comparably to our self-constructed ARM baselines. Remarkably, LLaDA 8B is competitive with strong LLMs like LLaMA3 8B in in-context learning and, after SFT, exhibits impressive instruction-following abilities in case studies such as multi-turn dialogue. Moreover, LLaDA addresses the reversal curse, surpassing GPT-4o in a reversal poem completion task. Our findings show the promise of diffusion models for language modeling at scale and challenge the common assumption that core LLM capabilities discussed above inherently depend on ARMs. Project page and codes: https://ml-gsai.github.io/LLaDA-demo/.