Yifan-Song793/GoodBadGreedy

The Good, The Bad, and The Greedy: Evaluation of LLMs Should Not Ignore Non-Determinism

19
/ 100
Experimental

This project helps evaluate how different generation strategies impact the performance of large language models (LLMs) in various tasks. It takes in LLM outputs generated using different decoding methods (like greedy or sampling) and benchmark evaluation results, then reveals how consistent an LLM's performance is and when one method outperforms another. This is for researchers, MLOps engineers, and product managers who are integrating LLMs and need to understand their reliability and optimal configuration.

No commits in the last 6 months.

Use this if you need to understand the variability of LLM outputs and how different generation settings (like greedy vs. sampling) affect performance on specific benchmarks for tasks like reasoning or code generation.

Not ideal if you are looking for a general-purpose LLM evaluation framework that doesn't focus on non-determinism or best-of-N sampling strategies.

LLM-evaluation natural-language-generation model-benchmarking AI-performance-analysis prompt-engineering
No License Stale 6m No Package No Dependents
Maintenance 0 / 25
Adoption 7 / 25
Maturity 8 / 25
Community 4 / 25

How are scores calculated?

Stars

30

Forks

1

Language

Python

License

Last pushed

Jul 17, 2024

Commits (30d)

0

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/llm-tools/Yifan-Song793/GoodBadGreedy"

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