Ecosyste.ms: Awesome

An open API service indexing awesome lists of open source software.

Awesome Lists | Featured Topics | Projects

https://github.com/amazon-science/RAGChecker

RAGChecker: A Fine-grained Framework For Diagnosing RAG
https://github.com/amazon-science/RAGChecker

Last synced: 6 days ago
JSON representation

RAGChecker: A Fine-grained Framework For Diagnosing RAG

Host: GitHub
URL: https://github.com/amazon-science/RAGChecker
Owner: amazon-science
License: apache-2.0
Created: 2024-06-24T13:53:07.000Z (6 months ago)
Default Branch: main
Last Pushed: 2024-08-20T10:51:58.000Z (4 months ago)
Last Synced: 2024-08-21T13:11:05.545Z (4 months ago)
Language: Python
Homepage:
Size: 2.99 MB
Stars: 186
Watchers: 5
Forks: 11
Open Issues: 5
Metadata Files:
- Readme: README.md
- Contributing: CONTRIBUTING.md
- License: LICENSE
- Code of conduct: CODE_OF_CONDUCT.md

Awesome Lists containing this project

awesome-production-machine-learning - RAGChecker - science/RAGChecker.svg?style=social) - RAGChecker is an advanced automatic evaluation framework designed to assess and diagnose Retrieval-Augmented Generation (RAG) systems. (Industry Strength Evaluation)
StarryDivineSky - amazon-science/RAGChecker - level entailment操作进行细粒度评估。基准数据集：全面的 RAG 基准数据集，包含涵盖 10 个领域的 4k 个问题。元评估：人工注释的偏好数据集，用于评估 RAGChecker 结果与人类判断的相关性。 RAGChecker 使开发人员和研究人员能够精确、深入地彻底评估、诊断和增强他们的 RAG 系统。 (A01_文本生成_文本对话 / 大语言对话模型及数据)

README

# RAGChecker: A Fine-grained Framework For Diagnosing RAG

RAGChecker Paper &nbsp&nbsp | &nbsp&nbsp Tutorial (English) &nbsp&nbsp ｜ &nbsp&nbsp 中文教程

RAGChecker is an advanced automatic evaluation framework designed to assess and diagnose Retrieval-Augmented Generation (RAG) systems. It provides a comprehensive suite of metrics and tools for in-depth analysis of RAG performance.

RefChecker Metrics

Figure: RAGChecker Metrics

## 🌟 Highlighted Features

- **Holistic Evaluation**: RAGChecker offers `Overall Metrics` for an assessment of the entire RAG pipeline.

- **Diagnostic Metrics**: `Diagnostic Retriever Metrics` for analyzing the retrieval component. `Diagnostic Generator Metrics` for evaluating the generation component. These metrics provide valuable insights for targeted improvements.

- **Fine-grained Evaluation**: Utilizes `claim-level entailment` operations for fine-grained evaluation.

- **Benchmark Dataset**: A comprehensive RAG benchmark dataset with 4k questions covering 10 domains (upcoming).

- **Meta-Evaluation**: A human-annotated preference dataset for evaluating the correlations of RAGChecker's results with human judgments.

RAGChecker empowers developers and researchers to thoroughly evaluate, diagnose, and enhance their RAG systems with precision and depth.

## 🔥 News
- [08/16/2024] RAGChecker paper is on arXiv: https://arxiv.org/pdf/2408.08067

## ❤️ Citation
RAGChecker paper: https://arxiv.org/pdf/2408.08067

If you use RAGChecker in your work, please cite us:
```bibtex
@misc{ru2024ragcheckerfinegrainedframeworkdiagnosing,
title={RAGChecker: A Fine-grained Framework for Diagnosing Retrieval-Augmented Generation},
author={Dongyu Ru and Lin Qiu and Xiangkun Hu and Tianhang Zhang and Peng Shi and Shuaichen Chang and Jiayang Cheng and Cunxiang Wang and Shichao Sun and Huanyu Li and Zizhao Zhang and Binjie Wang and Jiarong Jiang and Tong He and Zhiguo Wang and Pengfei Liu and Yue Zhang and Zheng Zhang},
year={2024},
eprint={2408.08067},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2408.08067},
}
```

## 🚀 Quick Start

### Setup Environment

```bash
pip install ragchecker
python -m spacy download en_core_web_sm
```

### Run the Checking Pipeline with CLI

Please process your own data with the same format as [examples/checking_inputs.json](./examples/checking_inputs.json). The only required annotation for each query is the `ground truth answer (gt_answer)`.

```json
{
"results": [
{
"query_id": "", # string
"query": "", # string
"gt_answer": "", # string
"response": "", # string
"retrieved_context": [ # a list of retrieved chunks by the retriever
{
"doc_id": "", # string, optional
"text": "" # string
},
...
]
},
...
]
}
```

If you are using AWS Bedrock version of Llama3 70B for the claim extractor and checker, use the following command to run the checking pipeline, the checking results as well as intermediate results will be saved to `--output_path`:

```bash
ragchecker-cli \
--input_path=examples/checking_inputs.json \
--output_path=examples/checking_outputs.json \
--extractor_name=bedrock/meta.llama3-1-70b-instruct-v1:0 \
--checker_name=bedrock/meta.llama3-1-70b-instruct-v1:0 \
--batch_size_extractor=64 \
--batch_size_checker=64 \
--metrics all_metrics \
# --disable_joint_check # uncomment this line for one-by-one checking, slower but slightly more accurate
```

Please refer to [RefChecker's guidance](https://github.com/amazon-science/RefChecker/tree/main?tab=readme-ov-file#choose-models-for-the-extractor-and-checker) for setting up the extractor and checker models.

It will output the values for the metrics like follows:

```json
Results for examples/checking_outputs.json:
{
"overall_metrics": {
"precision": 73.3,
"recall": 62.5,
"f1": 67.3
},
"retriever_metrics": {
"claim_recall": 61.4,
"context_precision": 87.5
},
"generator_metrics": {
"context_utilization": 87.5,
"noise_sensitivity_in_relevant": 22.5,
"noise_sensitivity_in_irrelevant": 0.0,
"hallucination": 4.2,
"self_knowledge": 25.0,
"faithfulness": 70.8
}
}
```

### Run the Checking Pipeline with Python
```python
from ragchecker import RAGResults, RAGChecker
from ragchecker.metrics import all_metrics

# initialize ragresults from json/dict
with open("examples/checking_inputs.json") as fp:
rag_results = RAGResults.from_json(fp.read())

# set-up the evaluator
evaluator = RAGChecker(
extractor_name="bedrock/meta.llama3-1-70b-instruct-v1:0",
checker_name="bedrock/meta.llama3-1-70b-instruct-v1:0",
batch_size_extractor=32,
batch_size_checker=32
)

# evaluate results with selected metrics or certain groups, e.g., retriever_metrics, generator_metrics, all_metrics
evaluator.evaluate(rag_results, all_metrics)
print(rag_results)

"""Output
RAGResults(
2 RAG results,
Metrics:
{
"overall_metrics": {
"precision": 76.4,
"recall": 62.5,
"f1": 68.3
},
"retriever_metrics": {
"claim_recall": 61.4,
"context_precision": 87.5
},
"generator_metrics": {
"context_utilization": 87.5,
"noise_sensitivity_in_relevant": 19.1,
"noise_sensitivity_in_irrelevant": 0.0,
"hallucination": 4.5,
"self_knowledge": 27.3,
"faithfulness": 68.2
}
}
)
"""
```

## Meta-Evaluation

Please refer to [data/meta_evaluation](./data/meta_evaluation/README.md) on meta-evaluation for the effectiveness of RAGChecker.

## Work with LlamaIndex

RAGChecker now integrates with LlamaIndex, providing a powerful evaluation tool for RAG applications built with LlamaIndex. For detailed instructions on how to use RAGChecker with LlamaIndex, please refer to the [LlamaIndex documentation on RAGChecker integration](https://docs.llamaindex.ai/en/latest/examples/evaluation/RAGChecker/). This integration allows LlamaIndex users to leverage RAGChecker's comprehensive metrics to evaluate and improve their RAG systems.

## Security

See [CONTRIBUTING](CONTRIBUTING.md#security-issue-notifications) for more information.

## License

This project is licensed under the Apache-2.0 License.