AI BriefingCoding with CodeQwen1.5
Key point
Qwen has released CodeQwen1.5-7B, an open-source code-specialized LLM trained on 3 trillion code tokens.
Details
Existing proprietary LLM-based coding assistants like GitHub Copilot face challenges around cost, privacy, security, and potential copyright infringement. As a transparent and accessible alternative, Qwen has introduced the open-source code LLM CodeQwen1.5-7B.
CodeQwen1.5-7B is built on the Qwen1.5 language model and was pretrained on approximately 3 trillion code-related tokens. It supports a total of 92 programming languages and is capable of handling long contexts of up to 64K tokens.
In terms of performance, CodeQwen1.5 demonstrates strong capabilities in basic code generation, long-context modeling, code editing, and SQL. Notably, on the HumanEval and MBPP benchmarks, it outperforms larger models despite having only 7B parameters, and it also ranks among the top open-source models on LiveCodeBench.
Furthermore, it has proven excellent programming ability not just in Python but across the 8 major languages covered by MultiPL-E, including Python, C++, Java, PHP, TypeScript, C#, Bash, and JavaScript.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.