Qwen: Qwen2.5 Coder 7B Instruct

Qwen2.5-Coder-7B-Instruct is a powerful 7B parameter instruction-tuned language model specifically engineered for a wide array of code-related tasks. Leveraging the advanced Qwen2.5 architecture, it incorporates state-of-the-art enhancements such as RoPE, SwiGLU, RMSNorm, and GQA attention. This model excels in code generation, complex code reasoning, and efficient bug fixing, making it an invaluable tool for developers. With support for an impressive context window of up to 128K tokens using YaRN-based extrapolation, Qwen2.5-Coder-7B-Instruct is trained on an extensive corpus of source code, synthetic data, and text-code grounding. This robust training ensures strong performance across various programming languages and facilitates agentic coding workflows. It offers a 32K token context window and 4K token max output. Pricing is $0.03/0.09 per 1M tokens (input/output) and it's available for FREE access. It also boasts strong compatibility with tools like vLLM for efficient deployment and is released under the permissive Apache 2.0 license.