6 citations · 19 across the 7 of their papers we have counts for
7 papers
Towards Compute-Optimal Many-Shot In-Context Learning
Shahriar Golchin, Yanfei Chen, Rujun Han +7
Long-context large language models (LLMs) are able to process inputs containing up to several million tokens. In the scope of in-context learning (ICL), this translates into using…
Speculative Knowledge Distillation: Bridging the Teacher-Student Gap Through Interleaved Sampling
Wenda Xu, Rujun Han, Zifeng Wang +7
Recent advances in knowledge distillation (KD) have enabled smaller student models to approach the performance of larger teacher models. However, popular methods such as supervised…
Don't Throw Away Data: Better Sequence Knowledge Distillation
Jun Wang, Eleftheria Briakou, Hamid Dadkhahi +3
A critical component in knowledge distillation is the means of coupling the teacher and student. The predominant sequence knowledge distillation method involves supervised learning…
On scalable oversight with weak LLMs judging strong LLMs
Zachary Kenton, Noah Y. Siegel, János Kramár +8
Scalable oversight protocols aim to enable humans to accurately supervise superhuman AI. In this paper we study debate, where two AI's compete to convince a judge; consultancy, whe…
Transformers Can Achieve Length Generalization But Not Robustly
Yongchao Zhou, Uri Alon, Xinyun Chen +3
Length generalization, defined as the ability to extrapolate from shorter training sequences to longer test ones, is a significant challenge for language models. This issue persist…
DistillSpec: Improving Speculative Decoding via Knowledge Distillation
Yongchao Zhou, Kaifeng Lyu, Ankit Singh Rawat +5
Speculative decoding (SD) accelerates large language model inference by employing a faster draft model for generating multiple tokens, which are then verified in parallel by the la…
On-Policy Distillation of Language Models: Learning from Self-Generated Mistakes
Rishabh Agarwal, Nino Vieillard, Yongchao Zhou +4
Knowledge distillation (KD) is widely used for compressing a teacher model to reduce its inference cost and memory footprint, by training a smaller student model. However, current…