Summary of Make Some Noise: Unlocking Language Model Parallel Inference Capability Through Noisy Training, by Yixuan Wang et al.
Make Some Noise: Unlocking Language Model Parallel Inference Capability through Noisy Training
by Yixuan Wang, Xianzhen Luo, Fuxuan Wei, Yijun Liu, Qingfu Zhu, Xuanyu Zhang, Qing Yang, Dongliang Xu, Wanxiang Che
First submitted to arxiv on: 25 Jun 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Machine Learning (cs.LG)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The proposed Make Some Noise (MSN) training framework is an alternative to supervised fine-tuning for large language models, introducing noise at the input to enhance parallel decoding capabilities without affecting original task performance. This approach enables significant improvements in inference speed, with a 2.3-2.7x increase in Spec-Bench and comparable acceleration ratios to the SOTA model. The MSN framework combines well with tree-based retrieval-augmented Jacobi (TR-Jacobi) decoding strategy for even faster inference speeds. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Large language models can decode draft tokens quickly, but existing methods require additional structure and training. A new approach called Make Some Noise (MSN) is proposed to help these models learn to denoise inputs. This makes the models much faster at processing tasks without changing how well they do their job. In fact, this new method makes some models 2-3 times faster! It also works with other techniques to make even more progress. |
Keywords
» Artificial intelligence » Fine tuning » Inference » Supervised
