English · 简体中文 · 日本語

👋 join us on Discord and WeChat

Changelog

• 2024/12/27: Project Initialization

Introduction

Dingo is a data quality evaluation tool that helps you automatically detect data quality issues in your datasets. Dingo provides a variety of built-in rules and model evaluation methods, and also supports custom evaluation methods. Dingo supports commonly used text datasets and multimodal datasets, including pre-training datasets, fine-tuning datasets, and evaluation datasets. In addition, Dingo supports multiple usage methods, including local CLI and SDK, making it easy to integrate into various evaluation platforms, such as OpenCompass.

Architecture Diagram

Quick Start

Installation

pip install dingo-python

Example Use Cases

1. Evaluate LLM chat data

from dingo.config.config import DynamicLLMConfig
from dingo.io.input.Data import Data
from dingo.model.llm.llm_text_quality_model_base import LLMTextQualityModelBase
from dingo.model.rule.rule_common import RuleEnterAndSpace

data = Data(
    data_id='123',
    prompt="hello, introduce the world",
    content="Hello! The world is a vast and diverse place, full of wonders, cultures, and incredible natural beauty."
)

def llm():
    LLMTextQualityModelBase.dynamic_config = DynamicLLMConfig(
        key='YOUR_API_KEY',
        api_url='https://api.openai.com/v1/chat/completions',
        model='gpt-4o',
    )
    res = LLMTextQualityModelBase.eval(data)
    print(res)


def rule():
    res = RuleEnterAndSpace().eval(data)
    print(res)

2. Evaluate Dataset

from dingo.io import InputArgs
from dingo.exec import Executor

# Evaluate a dataset from Hugging Face
input_data = {
    "eval_group": "sft",           # Rule set for SFT data
    "input_path": "tatsu-lab/alpaca", # Dataset from Hugging Face
    "data_format": "plaintext",    # Format: plaintext
    "save_data": True              # Save evaluation results
}

input_args = InputArgs(**input_data)
executor = Executor.exec_map["local"](input_args)
result = executor.execute()
print(result)

Command Line Interface

Evaluate with Rule Sets

python -m dingo.run.cli --input_path data.txt --dataset local -e sft --data_format plaintext --save_data True

Evaluate with LLM (e.g., GPT-4o)

python -m dingo.run.cli --input_path data.json --dataset local -e openai --data_format json --column_content text --custom_config config_gpt.json --save_data True

Example config_gpt.json:

{
  "llm_config": {
    "openai": {
      "model": "gpt-4o",
      "key": "YOUR_API_KEY",
      "api_url": "https://api.openai.com/v1/chat/completions"
    }
  }
}

GUI Visualization

After evaluation (with save_data=True), a frontend page will be automatically generated. To manually start the frontend:

python -m dingo.run.vsl --input output_directory

Where output_directory contains the evaluation results with a summary.json file.

Online Demo

Try Dingo on our online demo: (Hugging Face)🤗

Local Demo

Try Dingo in local:

cd app_gradio
python app.py

Google Colab Demo

Experience Dingo interactively with Google Colab notebook:

MCP Server

Dingo includes an experimental Model Context Protocol (MCP) server. For details on running the server and integrating it with clients like Cursor, please see the dedicated documentation:

English · 简体中文 · 日本語

Video Demonstration

To help you get started quickly with Dingo MCP, we've created a video walkthrough:

mcp_demo.mp4

This video demonstrates step-by-step how to use Dingo MCP server with Cursor.

Data Quality Metrics

Dingo provides comprehensive data quality assessment through both rule-based and prompt-based evaluation metrics. These metrics cover multiple quality dimensions including effectiveness, completeness, similarity, security, and more.

📊 View Complete Metrics Documentation →

Our evaluation system includes:

• Text Quality Assessment Metrics: Pre-training data quality evaluation using DataMan methodology and enhanced multi-dimensional assessment
• SFT Data Assessment Metrics: Honest, Helpful, Harmless evaluation for supervised fine-tuning data
• Classification Metrics: Topic categorization and content classification
• Multimodality Assessment Metrics: Image classification and relevance evaluation
• Rule-Based Quality Metrics: Automated quality checks using heuristic rules for effectiveness and similarity detection
• etc

Most metrics are backed by academic sources to ensure objectivity and scientific rigor.

Using LLM Assessment in Evaluation

To use these assessment prompts in your evaluations, specify them in your configuration:

input_data = {
    # Other parameters...
    "custom_config": {
        "prompt_list": ["QUALITY_BAD_SIMILARITY"],  # Specific prompt to use
        "llm_config": {
            "detect_text_quality": {  # LLM model to use
                "model": "gpt-4o",
                "key": "YOUR_API_KEY",
                "api_url": "https://api.openai.com/v1/chat/completions"
            }
        }
    }
}

You can customize these prompts to focus on specific quality dimensions or to adapt to particular domain requirements. When combined with appropriate LLM models, these prompts enable comprehensive evaluation of data quality across multiple dimensions.

Rule Groups

Dingo provides pre-configured rule groups for different types of datasets:

Group	Use Case	Example Rules
default	General text quality	RuleColonEnd, RuleContentNull, RuleDocRepeat, etc.
sft	Fine-tuning datasets	Rules from default plus RuleLineStartWithBulletpoint
pretrain	Pre-training datasets	Comprehensive set of 20+ rules including RuleAlphaWords, RuleCapitalWords, etc.

To use a specific rule group:

input_data = {
    "eval_group": "sft",  # Use "default", "sft", or "pretrain"
    # other parameters...
}

Feature Highlights

Multi-source & Multi-modal Support

• Data Sources: Local files, Hugging Face datasets, S3 storage
• Data Types: Pre-training, fine-tuning, and evaluation datasets
• Data Modalities: Text and image

Rule-based & Model-based Evaluation

• Built-in Rules: 20+ general heuristic evaluation rules
• LLM Integration: OpenAI, Kimi, and local models (e.g., Llama3)
• Custom Rules: Easily extend with your own rules and models
• Security Evaluation: Perspective API integration

Flexible Usage

• Interfaces: CLI and SDK options
• Integration: Easy integration with other platforms
• Execution Engines: Local and Spark

Comprehensive Reporting

• Quality Metrics: 7-dimensional quality assessment
• Traceability: Detailed reports for anomaly tracking

User Guide

Custom Rules, Prompts, and Models

If the built-in rules don't meet your requirements, you can create custom ones:

Custom Rule Example

from dingo.model import Model
from dingo.model.rule.base import BaseRule
from dingo.config.config import DynamicRuleConfig
from dingo.io import Data
from dingo.model.modelres import ModelRes

@Model.rule_register('QUALITY_BAD_RELEVANCE', ['default'])
class MyCustomRule(BaseRule):
    """Check for custom pattern in text"""

    dynamic_config = DynamicRuleConfig(pattern=r'your_pattern_here')

    @classmethod
    def eval(cls, input_data: Data) -> ModelRes:
        res = ModelRes()
        # Your rule implementation here
        return res

Custom LLM Integration

from dingo.model import Model
from dingo.model.llm.base_openai import BaseOpenAI

@Model.llm_register('my_custom_model')
class MyCustomModel(BaseOpenAI):
    # Custom implementation here
    pass

See more examples in:

• Register Rules
• Register Prompts
• Register Models

Execution Engines

Local Execution

from dingo.io import InputArgs
from dingo.exec import Executor

input_args = InputArgs(**input_data)
executor = Executor.exec_map["local"](input_args)
result = executor.execute()

# Get results
summary = executor.get_summary()        # Overall evaluation summary
bad_data = executor.get_bad_info_list() # List of problematic data
good_data = executor.get_good_info_list() # List of high-quality data

Spark Execution

from dingo.io import InputArgs
from dingo.exec import Executor
from pyspark.sql import SparkSession

# Initialize Spark
spark = SparkSession.builder.appName("Dingo").getOrCreate()
spark_rdd = spark.sparkContext.parallelize([...])  # Your data as Data objects

input_args = InputArgs(eval_group="default", save_data=True)
executor = Executor.exec_map["spark"](input_args, spark_session=spark, spark_rdd=spark_rdd)
result = executor.execute()

Evaluation Reports

After evaluation, Dingo generates:

1. Summary Report (summary.json): Overall metrics and scores
2. Detailed Reports: Specific issues for each rule violation

Report Description:

1. score: num_good / total
2. type_ratio: The count of type / total, such as: QUALITY_BAD_COMPLETENESS / total
3. name_ratio: The count of name / total, such as: QUALITY_BAD_COMPLETENESS-RuleColonEnd / total

Example summary:

{
    "task_id": "d6c922ec-981c-11ef-b723-7c10c9512fac",
    "task_name": "dingo",
    "eval_group": "default",
    "input_path": "test/data/test_local_jsonl.jsonl",
    "output_path": "outputs/d6c921ac-981c-11ef-b723-7c10c9512fac",
    "create_time": "20241101_144510",
    "score": 50.0,
    "num_good": 1,
    "num_bad": 1,
    "total": 2,
    "type_ratio": {
        "QUALITY_BAD_COMPLETENESS": 0.5,
        "QUALITY_BAD_RELEVANCE": 0.5
    },
    "name_ratio": {
        "QUALITY_BAD_COMPLETENESS-RuleColonEnd": 0.5,
        "QUALITY_BAD_RELEVANCE-RuleSpecialCharacter": 0.5
    }
}

Future Plans

• Richer graphic and text evaluation indicators
• Audio and video data modality evaluation
• Small model evaluation (fasttext, Qurating)
• Data diversity evaluation

Limitations

The current built-in detection rules and model methods focus on common data quality problems. For specialized evaluation needs, we recommend customizing detection rules.

Acknowledgments

• RedPajama-Data
• mlflow

Contribution

We appreciate all the contributors for their efforts to improve and enhance Dingo. Please refer to the Contribution Guide for guidance on contributing to the project.

License

This project uses the Apache 2.0 Open Source License.

This project uses fasttext for some functionality including language detection. fasttext is licensed under the MIT License, which is compatible with our Apache 2.0 license and provides flexibility for various usage scenarios.

Citation

If you find this project useful, please consider citing our tool:

@misc{dingo,
  title={Dingo: A Comprehensive Data Quality Evaluation Tool for Large Models},
  author={Dingo Contributors},
  howpublished={\url{https://github.com/DataEval/dingo}},
  year={2024}
}