Example

Visual Representation of Iterative Soft Prompt Refinement

The iterative refinement of soft prompts can be visualized as a multi-step process. In each step, the soft prompts from the previous iteration, denoted as σi\sigma_i, are processed by the model's Transformer layers along with the standard input embeddings (eie_i) and other vectors (ziz_i). The resulting hidden states (hih_i) from these layers are then used to compute an updated and more refined set of soft prompts, σi+1\sigma_{i+1}, for the subsequent step.

0

1

Updated 2025-10-10

Contributors are:

Who are from:

Tags

Ch.4 Alignment - Foundations of Large Language Models

Foundations of Large Language Models

Computing Sciences

Foundations of Large Language Models Course