Llama device

[ShawnXuan]

This PR introduces the ability to specify the target device (e.g., CUDA, NPU, XPU) for Llama inference. The changes allow users to select the desired device through configuration, improving flexibility across different hardware platforms.

• npu

  python projects/Llama/pipeline.py --device=npu --mode=huggingface --config_file=projects/Llama/configs/llama_config_npu.py

• xpu

  python projects/Llama/pipeline.py --device=xpu --mode=huggingface --config_file=projects/Llama/configs/llama_config_xpu.py

• cuda

Please update the projects/Llama/configs/llama_config.py file to configure the model path and tokenizer path.

python projects/Llama/pipeline.py