[MTL][Qwen] model fail RuntimeError: "normal_kernel_cpu" not implemented for 'Byte' about bigdl

[leonardozcm]

Sorry that for ipex-llm==2.5.0b20240421 && bigdl-core-xe-21 == 2.5.0b20240421, I can't reproduce this this issue on both arc and mtl(at least load_low_bit works fine)

(sgwhat-llm) D:\jinqiao\ipex-llm\python\llm\example\GPU\HF-Transformers-AutoModels\Model\qwen>python generate.py        
bin C:\Users\arda\miniconda3\envs\sgwhat-llm\lib\site-packages\bitsandbytes\libbitsandbytes_cpu.so
C:\Users\arda\miniconda3\envs\sgwhat-llm\lib\site-packages\bitsandbytes\cextension.py:34: UserWarning: The installed version of bitsandbytes was compiled without GPU support. 8-bit optimizers, 8-bit multiplication, and GPU quantization are unavailable.
  warn("The installed version of bitsandbytes was compiled without GPU support. "
function 'cadam32bit_grad_fp32' not found
C:\Users\arda\miniconda3\envs\sgwhat-llm\lib\site-packages\torchvision\io\image.py:13: UserWarning: Failed to load image Python extension: 'Could not find module 'C:\Users\arda\miniconda3\envs\sgwhat-llm\Lib\site-packages\torchvision\image.pyd' (or one of its dependencies). Try using the full path with constructor syntax.'If you don't plan on using image functionality from `torchvision.io`, you can ignore this warning. Otherwise, there might be something wrong with your environment. Did you have `libjpeg` or `libpng` installed before building `torchvision` from source?
  warn(
2024-04-22 13:48:15,416 - INFO - intel_extension_for_pytorch auto imported
Loading checkpoint shards: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 8/8 [00:00<00:00,  9.34it/s] 
2024-04-22 13:48:16,649 - INFO - Converting the current model to sym_int4 format......
-------------------- Prompt --------------------

<|im_start|>system
You are a helpful assistant.
<|im_end|>
<|im_start|>user
AI是什么？
<|im_end|>
<|im_start|>assistant

-------------------- Output --------------------

system
You are a helpful assistant.

user
AI是什么？

assistant
AI（人工智能）是指由计算机系统模拟、延伸和扩展人类智能的一门技术。它的目标是使计算机系统具有学习能力、推理能力和决策

[leonardozcm]

With passing fast_init=False, the model could be normally loaded.

model = AutoModelForCausalLM.load_low_bit(model_path, _fast_init=False, trust_remote_code=True, optimize_model=True).eval()

[jason-dai]

With passing fast_init=False, the model could be normally loaded.

model = AutoModelForCausalLM.load_low_bit(model_path, _fast_init=False, trust_remote_code=True, optimize_model=True).eval()

Shall we update our save/load example to explicitly add this parameter?

[leonardozcm]

Shall we update our save/load example to explicitly add this parameter?

Yes and I think we can do this in our load_low_bit API(so that our users will not be aware of the sight change and all examples will be ). For all saved low bit weights are already processed and need no initialized, this will not effect our function now.