CodeQwen1.5
About
CodeQwen1.5 is a family of open-source large language models (LLMs) designed specifically for code generation, leveraging the robust Qwen1.5 foundation model 14. These models are trained on an extensive dataset of approximately 3 trillion code tokens, supporting over 92 programming languages 145. Notably, their ability to process long contexts, accommodating up to 64,000 tokens, enables them to understand and generate code efficiently at the repository level 1. These models excel in various coding tasks, such as basic code generation, long-context modeling, code editing, and SQL querying, offering performance comparable to some closed-source models 1. The family includes both base models and chat-optimized versions like CodeQwen1.5-7B-Chat, which is crafted for interactive coding assistance 4. They incorporate advanced techniques like group query attention (GQA) for efficient inference 4.