'TinyGemmWeightQBitsTensor' object has no attribute '_scale'

[wikeeyang]

Hi, I use optimum-quanto v0.2.7 to quantized the https://huggingface.co/tencent/HunyuanImage-3.0 model and save to safetensors file, the command is "quantize(model, weights=qint4)", the method according to https://github.com/huggingface/optimum-quanto#quantization-workflow-for-vanilla-pytorch-models-low-level-api.

I can reload successed by script "load_quantized_model.py" which you can find under my HF repo https://huggingface.co/wikeeyang/Hunyuan-Image-30-Qint4.

use mmgp 3.6.4 want to save GPU memory, it will pop the below error message:

Create Meta model and Loading quantized weights to CPU...
************ Memory Management for the GPU Poor (mmgp 3.6.4) by DeepBeepMeep ************
Quantization of model 'transformer' started to format 'quanto.qint8'
No quantization to do as model is already quantized
Traceback (most recent call last):
File "/home/tkadm/HunyuanImage3/app_me.py", line 375, in
load_pipeline(args)
File "/home/tkadm/HunyuanImage3/app_me.py", line 40, in load_pipeline
hyi3_pipeline = HunyuanImage3AppPipeline(args)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/tkadm/HunyuanImage3/app/pipeline.py", line 83, in init
offload.profile( {"transformer":self.model.model}, profile_type.HighRAM_HighVRAM, verboseLevel = -1)
File "/home/tkadm/miniconda3/envs/video/lib/python3.11/site-packages/mmgp/offload.py", line 3074, in profile
return all(pipe_or_dict_of_modules, verboseLevel = verboseLevel, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/tkadm/miniconda3/envs/video/lib/python3.11/site-packages/mmgp/offload.py", line 2778, in all
dtype = p._scale.dtype
^^^^^^^^
AttributeError: 'TinyGemmWeightQBitsTensor' object has no attribute '_scale'

Please help to analyze the reason for this error, thank you a lot.