prometheus-eval

prometheus-eval/prometheus-eval

Evaluate your LLM's response with Prometheus and GPT4

1.1k
Stars
71
Forks
14
Open issues
2
Watchers
Python Apache-2.0Last pushed Apr 25, 2025

Overview

Prometheus-Eval is a Python-based tool for evaluating the performance of large language models using metrics from Prometheus and interactions with GPT-4.

Categories

Tags

Similar tools

Install

pip install prometheus-eval

README

Prometheus-Logo

🔥 Prometheus-Eval 🔥

arXiv Hugging Face Organization License PyPI version

⚡ A repository for evaluating LLMs in generation tasks 🚀 ⚡

Latest News 🔥

  • [2025/04] We release the latest iteration of Prometheus: M-Promethues (3B, 7B, & 14B)!

    • They outperform previous open LLM judges on multilingual meta-evaluation benchmarks (MM-Eval and M-RewardBench), and achieves exceptional results on literary translation evaluation.
    • The models also perform strongly in English, with the 7B and 14B models surpassing Prometheus 2 7B and 8x7B on RewardBench, respectively.
    • When used as judges at inference time, they significantly boost multilingual generation quality.
    • Checkout our paper, where we present extensive ablations to uncover the key factors behind effective multilingual judge training.
  • [2024/06] We release the BiGGen-Bench and Prometheus 2 BGB (8x7B)!

    • BiGGen-Bench features 9 core capabilities, 77 tasks, and 765 meticulously crafted instances, each with specific evaluation criteria.
    • We evaluated 103 frontier language models by 5 state-of-the-art evaluator language models and analyzed the findings in our paper.
    • We continually trained Prometheus 2 8x7B on BiGGen-Bench evaluation trace and built our most capable evaluator LM Prometheus 2 BGB, even surpassing Claude-3-Opus on absolute grading tasks.
    • Checkout our dataset, evaluation results, leaderboard, interactive report, and the code!
  • [2024/05] We release Prometheus 2 (7B & 8x7B) models!

    • Prometheus 2 (8x7B) is an open-source state-of-the-art evaluator language model!
      • Compared to Prometheus 1 (13B), Prometheus 2 (8x7B) shows improved evaluation performances & supports assessing in pairwise ranking (relative grading) formats as well!
      • It achieves a Pearson correlation of 0.6 to 0.7 with GPT-4-1106 on a 5-point Likert scale across multiple direct assessment benchmarks, including VicunaBench, MT-Bench, and FLASK.
      • It also scores a 72% to 85% agreement with human judgments across multiple pairwise ranking benchmarks, including HHH Alignment, [MT Bench Human Judgment](