Closed 00010110 closed 2 months ago
Both models possess the capability for code completion, yet they are best utilized through distinct approaches.
For the base model, we advise employing the FIM (Fill-in-the-Middle) format incorporating specialized tokens to enhance its performance.
Conversely, the instruct model is optimally utilized by adopting a natural language format, thereby leveraging its understanding of instructions to facilitate code autocompletion tasks effectively.
Qwen2.5-Coder-Instruct版本是否有代码补全能力?