Turning Up the Heat: Min-p Sampling for Creative and Coherent LLM Outputs
This addresses the issue of incoherent or repetitive outputs in LLM text generation, particularly at higher temperatures, for users of open-source frameworks, though it is incremental as it builds on existing sampling methods.
The paper tackled the problem of balancing quality and diversity in text generation from large language models by proposing min-p sampling, a dynamic truncation method that adjusts the sampling threshold based on model confidence, resulting in improved quality and diversity across benchmarks and model sizes, with human evaluations showing a clear preference for min-p sampling.
Large Language Models (LLMs) generate text by sampling the next token from a probability distribution over the vocabulary at each decoding step. Popular sampling methods like top-p (nucleus sampling) often struggle to balance quality and diversity, especially at higher temperatures which lead to incoherent or repetitive outputs. We propose min-p sampling, a dynamic truncation method that adjusts the sampling threshold based on the model's confidence by using the top token's probability as a scaling factor. Our experiments on benchmarks including GPQA, GSM8K, and AlpacaEval Creative Writing show that min-p sampling improves both the quality and diversity of generated text across different model families (Mistral and Llama 3) and model sizes (1B to 123B parameters), especially at higher temperatures. Human evaluations further show a clear preference for min-p sampling, in both text quality and creativity. Min-p sampling has been adopted by popular open-source LLM frameworks, including Hugging Face Transformers, VLLM, and many others, highlighting its considerable impact on improving text generation quality.