LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
Why it matters
A deep technical survey of how to steer LLMA large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.Full definition → outputs — covering guided decoding, prompt tuning, and unlikelihood training — for anyone who needs finer control over what an LLM generates than raw prompting provides.
Key quotes
“By adopting different decoding methods, we can place restrictions or preferences on the sampling process to alter the generated samples without modifying any model weights.”
Lilian Weng
“However, maximization-based decoding does not guarantee high-quality generation.”
Lilian Weng
“Guided decoding essentially runs a more expensive beam search where the sampling probability distribution is altered by side information about human preferences.”
Lilian Weng
“Both AutoPrompt and Prefix-Tuning were found to outperform fine-tuning in the regime where the training dataset is small (i.e. $10^2-10^3$ samples).”
Lilian Weng
“Prompt tuning produces competitive results as model fine-tuning when the model gets large (billions of parameters and up). This result is especially interesting given that large models are expensive to fine-tune and execute at inference time.”
articlePrompt EngineeringLilian Weng
articleBeyond Prompt Engineering: A Systematic Analysis of Prompt Lexical Sensitivity and Its Impacts on QualityQipeng Xie, Zi Liang, Jiafei Wu, Yufei Chen, Weizheng Wang, Wenao Ma, Zhong Ming, Haiqin Yang, Kaishun Wu
articleTraining language models to follow instructions with human feedbackLong Ouyang et al.