Verbalized Sampling
What it is
Verbalized Sampling is a training-free prompting strategy that mitigates mode collapse in Large Language Models by requesting multiple responses with associated probabilities. It allows users to sample from the 'tails' of these distributions to increase output diversity without retraining the model.
The problem it solves
Mode collapse in LLMs, where models tend to produce repetitive or low-diversity responses for the same prompt.
What you could use it for
- Creative writing with higher variety
- Synthetic data generation
- Dialogue simulation
- Open-ended QA
- Text-to-image prompt diversity
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.