Model Overview Card for Qwen 2.5 7B Instruct Turbo

Basic Information

• Model Name: Qwen 2.5 7B Instruct Turbo
• Developer/Creator: Alibaba Group
• Release Date: September 18, 2024
• Version: 2.5
• Model Type: Text

Description

Overview:

The Qwen 2.5 7B Instruct model is a cutting-edge large language model designed to understand and generate text based on specific instructions. It excels in various tasks, including coding, mathematical problem-solving, and generating structured outputs.

Key Features:

• Supports long-context inputs up to 131,072 tokens.
• Generates outputs of up to 8,192 tokens.
• Enhanced instruction-following capabilities.
• Multilingual support for over 29 languages, including English, Chinese, Spanish, and more.
• Improved performance in coding and mathematics compared to previous versions.

Intended Use:

This model is intended for software developers, researchers, and businesses looking to leverage advanced natural language processing capabilities in applications such as:

• Automated content generation (articles, reports).
• Coding assistance (code generation, debugging).
• AI-driven chatbots and virtual assistants.

Language Support:

Qwen 2.5 supports multiple languages, making it versatile for global applications.

Technical Details

Architecture:

Qwen 2.5 utilizes a Transformer architecture with enhancements like RoPE (Rotary Positional Embedding), SwiGLU activation functions, RMSNorm normalization, and Attention QKV bias. It consists of 28 layers and 28 attention heads for query processing.

Training Data:

The model was trained on an extensive dataset comprising over 18 trillion tokens, sourced from diverse domains such as books, websites, and programming repositories. This broad dataset enhances its understanding of various topics.

Data Source and Size:

The training data includes a rich mix of text types and programming languages, ensuring the model's robustness and adaptability across different contexts.

Knowledge Cutoff:

The model's knowledge is current as of October 2024.

Diversity and Bias:

Efforts were made to ensure the training data is diverse to reduce biases. However, like all AI models, it may still reflect some inherent biases present in the data.

Performance Metrics

Key performance metrics for Qwen 2.5 7B Instruct include:

• Accuracy: Achieved an MMLU score of approximately 74.2, indicating strong performance in language understanding tasks.
• Speed: Optimized for fast inference, making it suitable for real-time applications.
• Robustness: Demonstrates high adaptability across diverse inputs and maintains performance even with complex queries.

Comparison to Other Models

Usage

Code Samples:

‍

Ethical Guidelines

The development of Qwen 2.5 adheres to ethical AI principles, emphasizing transparency, fairness, and accountability in its applications. Users are encouraged to consider these guidelines when deploying the model for various tasks.

Licensing

LThe Qwen 2.5 models are available under the Apache 2.0 License for commercial and non-commercial use.
‍