How to use custom dataloder in QAT?

huangqiu15444 commented 6 days ago

🐛 Describe the bug

TypeError: conv2d() received an invalid combination of arguments - got (dict, Parameter, Parameter, tuple, tuple, tuple, int), but expected one of:

(Tensor input, Tensor weight, Tensor bias, tuple of ints stride, tuple of ints padding, tuple of ints dilation, int groups) didn't match because some of the arguments have invalid types: (dict, Parameter, Parameter, tuple of (int, int), tuple of (int, int), tuple of (int, int), int)
(Tensor input, Tensor weight, Tensor bias, tuple of ints stride, str padding, tuple of ints dilation, int groups) didn't match because some of the arguments have invalid types: (dict, Parameter, Parameter, tuple of (int, int), tuple of (int, int), tuple of (int, int), int)

Is there is way to use a custom dataloader in the same way as “def transform_fn(dataitem): images, = data_item return images”?

Environment

about-time==4.2.1 albumentations==1.3.1 alive-progress==3.1.5 attrs==24.2.0 autograd==1.7.0 certifi==2024.8.30 chardet==5.2.0 charset-normalizer==3.3.2 cma==3.2.2 colorama==0.4.6 coloredlogs==15.0.1 contourpy==1.1.1 cycler==0.12.1 Deprecated==1.2.14 dill==0.3.8 einops==0.8.0 filelock==3.15.4 flatbuffers==24.3.25 fonttools==4.53.1 fsspec==2024.6.1 grapheme==0.6.0 humanfriendly==10.0 idna==3.8 imageio==2.35.1 importlib_resources==6.4.4 Jinja2==3.1.4 joblib==1.4.2 jsonschema==4.23.0 jsonschema-specifications==2023.12.1 jstyleson==0.0.2 kiwisolver==1.4.5 lazy_loader==0.4 lightning-utilities==0.11.6 markdown-it-py==3.0.0 MarkupSafe==2.1.5 matplotlib==3.7.5 mdurl==0.1.2 mpmath==1.3.0 natsort==8.4.0 networkx==3.1 ninja==1.11.1.1 nncf==2.12.0 numpy==1.24.4 onnx==1.16.2 onnxoptimizer==0.3.13 onnxruntime==1.19.0 onnxsim==0.4.36 opencv-python==4.10.0.84 opencv-python-headless==4.10.0.84 openvino==2024.3.0 openvino-telemetry==2024.1.0 packaging==24.1 pandas==2.0.3 pillow==10.4.0 pkgutil_resolve_name==1.3.10 protobuf==5.28.0 psutil==6.0.0 pyclipper==1.3.0.post5 pydot==2.0.0 Pygments==2.18.0 pymoo==0.6.1.2 pyparsing==3.1.4 pyreadline3==3.4.1 python-dateutil==2.9.0.post0 pytz==2024.1 PyWavelets==1.4.1 PyYAML==6.0.2 qudida==0.0.4 referencing==0.35.1 requests==2.32.3 rich==13.8.0 rpds-py==0.20.0 scikit-image==0.21.0 scikit-learn==1.3.2 scipy==1.10.1 seaborn==0.13.2 shapely==2.0.6 six==1.16.0 sympy==1.13.2 tabulate==0.9.0 tensorboardX==2.6.2.2 threadpoolctl==3.5.0 tifffile==2023.7.10 torch==2.2.2+cu121 torchaudio==2.2.2+cu121 torchmetrics==1.4.1 torchsummary==1.5.1 torchvision==0.17.2+cu121 tqdm==4.66.5 typing_extensions==4.12.2 tzdata==2024.1 urllib3==2.2.2 wrapt==1.16.0 zipp==3.20.1

Minimal Reproducible Example

nncf_config = NNCFConfig.from_json("int8_per_tensor.json") nncf_config = register_default_init_args(nncf_config, self.train_loader) compression_ctrl, quantized_model = create_compressed_model(self.model, nncf_config)

Are you going to submit a PR?

[X] Yes I'd like to help by submitting a PR!

alexsu52 commented 5 days ago

Hello @huangqiu15444,

create_compressed_model will be deprecated in the next release. Please use nncf.quantize for QAT. Here is example: https://github.com/openvinotoolkit/nncf/tree/develop/examples/quantization_aware_training/torch/resnet18

huangqiu15444 commented 5 days ago

Hello @huangqiu15444,

create_compressed_model will be deprecated in the next release. Please use nncf.quantize for QAT. Here is example: https://github.com/openvinotoolkit/nncf/tree/develop/examples/quantization_aware_training/torch/resnet18

Thank you for your reply. I would like to know how to configure parameters such as ‘per_channel‘ in nncf.quantize(model, quantization_dataset).

alexsu52 commented 4 days ago

You can manage quantization parameters via advanced_parameters argument of nncf.quantize: https://openvinotoolkit.github.io/nncf/autoapi/nncf/quantization/advanced_parameters/index.html

openvinotoolkit / nncf