Support for Qwen2 (Llama2 based)

[bevhanno]

Adding Qwen2 model module that hast been tested with Qwen2-7B.

The Qwen2 module is based on the Llama2 module and differs in following points:

• the QwenAttention module has bias=True in the QKV projection
• the QKV projection bias needs to be loaded in the load_weights function of the model

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.