Lookahead Unmasking Elicits Accurate Decoding in Diffusion Language Models
arxiv.org · 7,526 words · saved by 1 readers
N/A
Preprint L OOKAHEAD U NMASKING E LICITS ACCURATE D ECODING IN D IFFUSION L ANGUAGE M ODELS Sanghyun Lee1∗, Seungryong Kim1 , Jongho Park2 , Dongmin Park3† 1 KAIST 2 University of California, Berkeley 3 KRAFTON {lsh83210,seungryong.kim}@kaist.ac.kr jjhpark@berkeley.edu dongmin.park@krafton.com…
related reading
- Large Language Diffusion Modelsarxiv.org
- 2409.02908arxiv.org
- Esoteric Language Modelsarxiv.org
- 2406.07524arxiv.org
- Fast Inference from Transformers via Speculative Decodingarxiv.org
- 2503.09573arxiv.org
- Self-Adapting Language Modelsarxiv.org
- [2402.12374] Sequoia: Scalable, Robust, and Hardware-aware Speculative Decodingarxiv.org
- A Survey on Sparse Autoencoders: Interpreting the Internal Mechanisms of Large Language Modelsarxiv.org
- Sparse Autoencoders Find Highly Interpretable Features in Language Modelsarxiv.org
- Diffusion Language Model Inference with Monte Carlo Tree Searcharxiv.org
- Accelerating Diffusion LLMs via Adaptive Parallel Decodingarxiv.org