Qwen2: can not run with the latest Qwen2 models

[zensh]

I am trying to run Qwen2 example with the latest Qwen2 0.5B model, but get a Error: cannot find tensor lm_head.weight.

https://huggingface.co/Qwen/Qwen2-0.5B

[LaurentMazare]

Ah thanks for reporting this, indeed there is a small difference in the new model architecture, I'm minting #2257 to support this and you should be able to use the new model with --model 2-0.5b.

[zensh]

@LaurentMazare Thank you very much, it’s now running. However, the results from the 0.5B model are a bit strange, which might be due to other issues.

[deadash]

7b is incorrect，but cpu is ok.