Extracting Latent Steering Vectors from Pretrained Language Models. (arXiv:2205.05124v1 [cs.CL])

Web: http://arxiv.org/abs/2205.05124

Prior work on controllable text generation has focused on learning how to
control language models through trainable decoding, smart-prompt design, or
fine-tuning based on a desired objective. We hypothesize that the information
needed to steer the model to generate a target sentence is already encoded
within the model. Accordingly, we explore a different approach altogether:
extracting latent vectors directly from pretrained language model decoders
without fine-tuning. Experiments show that there exist steering vectors, which,
when added to the hidden states …

arxiv language language models models