Zobrazeno 1 - 6
of 6
pro vyhledávání: '"Neo, Clement"'
Autor:
Haimes, Jacob, Wenner, Cenny, Thaman, Kunvar, Tashev, Vassil, Neo, Clement, Kran, Esben, Schreiber, Jason
The training data for many Large Language Models (LLMs) is contaminated with test data. This means that public benchmarks used to assess LLMs are compromised, suggesting a performance gap between benchmark scores and actual capabilities. Ideally, a p
Externí odkaz:
http://arxiv.org/abs/2410.09247
Vision-Language Models (VLMs) are powerful tools for processing and understanding text and images. We study the processing of visual tokens in the language model component of LLaVA, a prominent VLM. Our approach focuses on analyzing the localization
Externí odkaz:
http://arxiv.org/abs/2410.07149
Large Language Models (LLMs) generate text by sampling the next token from a probability distribution over the vocabulary at each decoding step. However, popular sampling methods like top-p (nucleus sampling) often struggle to balance quality and div
Externí odkaz:
http://arxiv.org/abs/2407.01082
Understanding the inner workings of large language models (LLMs) is crucial for advancing their theoretical foundations and real-world applications. While the attention mechanism and multi-layer perceptrons (MLPs) have been studied independently, the
Externí odkaz:
http://arxiv.org/abs/2402.15055
Language Models (LMs) are increasingly used for a wide range of prediction tasks, but their training can often neglect rare edge cases, reducing their reliability. Here, we define a stringent standard of trustworthiness whereby the task algorithm and
Externí odkaz:
http://arxiv.org/abs/2402.02619
Autor:
Marks, Luke, Abdullah, Amir, Neo, Clement, Arike, Rauno, Krueger, David, Torr, Philip, Barez, Fazl
Reinforcement learning from human feedback (RLHF) is widely used to train large language models (LLMs). However, it is unclear whether LLMs accurately learn the underlying preferences in human feedback data. We coin the term \textit{Learned Feedback
Externí odkaz:
http://arxiv.org/abs/2310.08164