ZhouYuxuanYX/Benchmarking-and-Guiding-Adaptive-Sampling-Decoding-for-LLMs

This is the official implementation of our ACL 2025 Main paper "Balancing Diversity and Risk in LLM Sampling".

25
/ 100
Experimental

When you're building or using large language models for open-ended text generation, this project helps you choose the best sampling method to get diverse and stable outputs. It provides an evaluation framework and benchmarks different methods, showing what goes in (your model's predicted word probabilities) and what comes out (guidance on which sampling method and parameters to use). This is for AI researchers, ML engineers, or anyone working with LLMs who needs to fine-tune text generation for specific creative or practical applications.

Use this if you need to optimize the quality of text generated by large language models, specifically balancing creativity (diversity) with reliability (stability) to avoid nonsensical or irrelevant outputs.

Not ideal if you are a general user of LLMs and don't need to dive into the technical details of sampling methods or fine-tune their behavior.

large-language-models text-generation natural-language-processing AI-model-evaluation machine-learning-engineering
No License No Package No Dependents
Maintenance 6 / 25
Adoption 6 / 25
Maturity 8 / 25
Community 5 / 25

How are scores calculated?

Stars

17

Forks

1

Language

Python

License

Last pushed

Oct 16, 2025

Commits (30d)

0

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/transformers/ZhouYuxuanYX/Benchmarking-and-Guiding-Adaptive-Sampling-Decoding-for-LLMs"

Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.