Qwen2.5-1.5B-Instruct report error

[Chaofan34]

use model Qwen2.5-1.5B-Instruct is report error

but Qwen3-0.6B is success

[jiahe7ay]

use model Qwen2.5-1.5B-Instruct is report error

but Qwen3-0.6B is success

@Chaofan34 This is because the model module in nano-vllm only implements Qwen3, but there are some minor differences between Qwen3 and Qwen2.5. So the reason for your error is that you’re using the Qwen3 module to load Qwen2.5 weights, causing some weights to not find corresponding modules. You can refer to the Qwen3 implementation in the project and the Qwen2.5 implementation on Hugging Face to create your own Qwen2.5 code for adaptation.

[Zlzzzupup]

I've submitted a pull request to add the Qwen2 model. You can review the PR when you have a moment. #70