6 papers
POROver: Improving Safety and Reducing Overrefusal in Large Language Models with Overgeneration and Preference Optimization
Batuhan K. Karaman, Ishmam Zabir, Alon Benhaim +3
Achieving both high safety and high usefulness simultaneously in large language models has become a critical challenge in recent years.Models often exhibit unsafe behavior or adopt…
A Practical Analysis of Human Alignment with *PO
Kian Ahrabian, Xihui Lin, Barun Patra +4
At the forefront of state-of-the-art human alignment methods are preference optimization methods (*PO). Prior research has often concentrated on identifying the best-performing met…
Scaling Optimal LR Across Token Horizons
Johan Bjorck, Alon Benhaim, Vishrav Chaudhary +2
State-of-the-art LLMs are powered by scaling -- scaling model size, dataset size and cluster size. It is economically infeasible to extensively tune hyperparameter for the largest…
S2-Attention: Hardware-Aware Context Sharding Among Attention Heads
Xihui Lin, Yunan Zhang, Suyu Ge +5
Sparse attention, which selectively attends to a subset of tokens in the context was supposed to be efficient. However, its theoretical reduction in FLOPs has rarely translated int…
Scaling Laws for Multilingual Language Models
Yifei He, Alon Benhaim, Barun Patra +6
We propose a novel scaling law for general-purpose decoder-only language models (LMs) trained on multilingual data, tackling the problem of balancing languages during multilingual…
On The Adaptation of Unlimiformer for Decoder-Only Transformers
Kian Ahrabian, Alon Benhaim, Barun Patra +3
One of the prominent issues stifling the current generation of large language models is their limited context length. Recent proprietary models such as GPT-4 and Claude 2 have intr…