Quantizing model in per-tensor way does't work

[WanliZhong]

Hi, I try to quantize all the parameters of my model in a per_tensor way. But the result shows that there are still some layers not in per_tensor way as below image shown.

The config file is following:

version: 1.0

model:                                               # mandatory. used to specify model specific information.
  name: mp_handpose
  framework: onnxrt_qlinearops                       # mandatory. supported values are tensorflow, pytorch, pytorch_ipex, onnxrt_integer, onnxrt_qlinear or mxnet; allow new framework backend extension.

quantization:                                        # optional. tuning constraints on model-wise for advance user to reduce tuning space.
  approach: post_training_static_quant               # optional. default value is post_training_static_quant.                          
  calibration:
    dataloader:
      batch_size: 1
      dataset:
        dummy:
          shape: [1, 256, 256, 3]
          low: -1.0
          high: 1.0
          dtype: float32
          label: True

  model_wise:                                        # optional. tuning constraints on model-wise for advance user to reduce tuning space.
    weight:
      granularity: per_tensor
      scheme: asym
      dtype: int8
      algorithm: minmax
    activation:
      granularity: per_tensor
      scheme: asym
      dtype: int8, fp32
      algorithm: minmax, kl

tuning:
  accuracy_criterion:
    relative:  0.02                                  # optional. default value is relative, other value is absolute. this example allows relative accuracy loss: 1%.
  exit_policy:
    timeout: 0                                       # optional. tuning timeout (seconds). default value is 0 which means early stop. combine with max_trials field to decide when to exit.
  random_seed: 9527                                  # optional. random seed for deterministic tuning.

The model I use is https://github.com/WanliZhong/opencv_zoo/blob/handpose_tracker/models/handpose_estimation_mediapipe/handpose_estimation_mediapipe_2022may.onnx

Thanks!

[mengniwang95]

Hi, I try it on my local env and this is the result:

please try

  model_wise:                                        # optional. tuning constraints on model-wise for advance user to reduce tuning space.
    weight:
      granularity: per_tensor
      scheme: asym
      dtype: int8
      algorithm: minmax
    activation:
      granularity: per_tensor
      scheme: asym
      dtype: int8
      algorithm: minmax

[WanliZhong]

Thanks!
I realized that the version I was using was too old and I forgot to install onnxruntime-extensions. Now I got the result model I wanted.