Solution for quantize-lm does not work on macOS

[BSVogler]

Hi! I am currently looking into the quantization on the task easy, quantize-lm, with torch and had some difficulties. The solution has the same issue.

On macOS on ARM with torch 2.7.1. the torch.backends.quantized.engine is by default set to none. So it will complain about no engine beeing set.

it can be set with

if 'qnnpack' in torch.backends.quantized.supported_engines:
    torch.backends.quantized.engine = 'qnnpack'

Next it fails in this line with

quantized_model.load_state_dict(torch.load("quantized_language_model.pth", weights_only=True))

UnpicklingError: Weights only load failed. This file can still be loaded, to do so you have two options, do those steps only if you trust the source of the checkpoint.
(1) In PyTorch 2.6, we changed the default value of the weights_only argument in torch.load from False to True. Re-running torch.load with weights_only set to False will likely succeed, but it can result in arbitrary code execution. Do it only if you got the file from a trusted source.
(2) Alternatively, to load with weights_only=True please check the recommended steps in the following error message.
WeightsUnpickler error: Unsupported global: GLOBAL torch.ScriptObject was not an allowed global by default. Please use torch.serialization.add_safe_globals([torch.ScriptObject]) or the torch.serialization.safe_globals([torch.ScriptObject]) context manager to allowlist this global if you trust this class/function.

It works when set to false but not sure if this is following best practices.