加载中...

Data Methodology

How we collect and organize LLM and benchmark data

Last updated: 2025-01-20

DataLearnerAI is committed to providing accurate and reliable AI model information. This page explains our data collection process and source priorities.

Data Collection Principles

We follow these principles to ensure data accuracy and authority:

1Prioritize officially published data to ensure authoritative information
2Label data sources for discrepancies so users can verify
3Regularly update data to reflect the latest model releases and benchmarks
4Maintain transparency in data collection and accept user feedback

Data Source Priority

We collect data according to the following priority:

First Priority

Official Published Data

Data directly from model publishers, including:

Official GitHub repository README and documentation
Official Hugging Face model cards
Publisher website product pages and tech blogs
Academic papers (arXiv, ACL, NeurIPS, etc.)

Second Priority

Authoritative Benchmarks

Official results from renowned benchmarks:

Official leaderboards for MMLU, GSM8K, HumanEval, etc.
Community-maintained evaluations like Open LLM Leaderboard
Human evaluations like LMSYS Chatbot Arena

Third Priority

Third-Party Evaluators

Data from reputable independent evaluation organizations:

Artificial Analysis model performance data
Other professional AI evaluation websites
Verified community reproduction results

Handling Data Conflicts

When data from different sources conflict, we apply these strategies:

Prioritize Official Data

Officially published data has the highest authority.

Label Data Sources

Key data points include source references for user verification.

Preserve Multiple Versions

When differences are significant, we may show data from multiple sources.

Continuous Updates

We update data promptly as new information becomes available.

Data Types

Data Type	Description	Primary Source
Model Basic Info	Parameter count, context length, release date, licenses, etc.	Primarily from official GitHub/Hugging Face and papers
Benchmark Scores	Evaluation results from various benchmarks	Official published results preferred, then benchmark leaderboards
API Pricing	Model API pricing information	From official pricing pages, updated regularly
Performance Metrics	Inference speed, throughput, and other performance data	Official data or evaluators like Artificial Analysis

Feedback & Corrections

If you find any data errors or have authoritative source suggestions, please use the contact options in the footer to reach us.