Qwen2.5-Coder: Unleash the Power of AI-Driven Code Generation
Welcome to Qwen2.5-Coder, the advanced code generation solution developed by the Qwen team at Alibaba Cloud. Building upon the robust foundation of open-source code large language models (CodeLLMs), this series is designed to empower developers with highly efficient coding capabilities, featuring Powerful, Diverse, and Practical solutions for various programming needs.

Key Features:

• Powerful Performance:
  Qwen2.5-Coder-32B-Instruct has achieved state-of-the-art performance, rivaling GPT-4o in coding tasks, while also showcasing strong general and mathematical skills.
• Diverse Model Sizes:
  The Qwen2.5-Coder series introduces a total of six mainstream model sizes, including 0.5B, 3B, 14B, and 32B, catering to the varied demands of developers across different project scales.
• Practical Applications:
  This model excels in two main scenarios: as a code assistant and a tool for practical coding deployments.

Model Specifications:

1. Extended Context Length: With the ability to understand and generate content with a context length of up to 128K tokens, Qwen2.5-Coder can process extensive codebases effectively.

2. Multi-language Support: This model supports 92 programming languages, including but not limited to Python, Java, C++, JavaScript, and many others.

   supported_languages = [  
       'ada', 'c', 'python', 'java', 'ruby', 'rust', 'javascript',  
       'golang', 'html', 'css', 'sql'  
   ]  
   

3. Enhanced Mathematical and General Reasoning: Qwen2.5-Coder retains robust capabilities in mathematical reasoning and other complex general tasks.

Quick Start:

To start interacting with Qwen2.5-Coder, you can initiate a chat through the following Python snippet:

from transformers import AutoModelForCausalLM, AutoTokenizer  
  
model_name = "Qwen/Qwen2.5-Coder-32B-Instruct"  
model = AutoModelForCausalLM.from_pretrained(model_name)  
tokenizer = AutoTokenizer.from_pretrained(model_name)  
   
prompt = "write a quick sort algorithm."  
messages = [  
    {"role": "system", "content": "You are Qwen, created by Alibaba Cloud. You are a helpful assistant."},  
    {"role": "user", "content": prompt}  
]  
text = tokenizer.apply_chat_template(messages).  
model_inputs = tokenizer(text, return_tensors="pt")  

generated_ids = model.generate(**model_inputs, max_new_tokens=512)  
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]  

This example demonstrates how to leverage Qwen2.5-Coder for coding tasks, showcasing its ability to generate code fluently.

Advanced Features:

Qwen2.5-Coder also supports efficient processing for extensive inputs, and file-level code completion for seamless code organization. For instance, utilizing special tokens such as <|repo_name|>, and <|file_sep|> aids in managing repository-level tasks effectively.

Performance and Research:

For detailed performance metrics and research insights, please refer to the Qwen2.5-Coder Technical Report. Stay updated through our blog and join our community on Discord for discussions about implementation and use cases.

Join Us:

If you find the Qwen2.5-Coder capabilities beneficial, please star the repository and consider citing us in your work. Together, let's foster an innovative coding experience powered by AI!