def get_predefined_configuration(
data: Dict[str, Any], ) -> Union[Dict[str, Any], List[Dict[str, Any]]]:
"""Get configuration."""
from lpot.ux.utils.utils import get_framework_from_path, get_predefined_config_path
from lpot.ux.utils.workload.config import Config
model_path = data.get("model_path", "")
if not os.path.isfile(model_path):
raise ClientErrorException(
f"Could not find model in specified path: {model_path}.", )
model_name = Path(model_path).stem
domain = data.get("domain", None)
if not domain:
raise ClientErrorException("Domain is not defined!")
framework = get_framework_from_path(model_path)
if framework is None:
raise ClientErrorException(
f"Could not find framework for specified model {model_name} in path {model_path}.",
)
config = Config()
predefined_config_path = get_predefined_config_path(framework, domain)
config.load(predefined_config_path)
return {
"config": config.serialize(),
"framework": framework,
"name": model_name,
"domain": domain,
}
def test_load(self) -> None:
"""Test load."""
config = Config()
read_yaml = yaml.dump(self.predefined_config, sort_keys=False)
with patch(
"lpot.ux.utils.workload.config.open",
mock_open(read_data=read_yaml),
) as mocked_open:
config.load("path to yaml file")
mocked_open.assert_called_once_with("path to yaml file")
expected = Config(self.predefined_config)
self.assertEqual(expected.serialize(), config.serialize())
def test_dump(self) -> None:
"""Test dump."""
config = Config(self.predefined_config)
expected_yaml = yaml.dump(
config.serialize(),
indent=4,
default_flow_style=None,
sort_keys=False,
)
with patch("lpot.ux.utils.workload.config.open",
mock_open()) as mocked_open:
config.dump("path to yaml file")
mocked_open.assert_called_once_with("path to yaml file", "w")
mocked_open().write.assert_called_once_with(expected_yaml)
def test_config_serializer(self) -> None:
"""Test Config serializer."""
config = Config(predefined_config)
result = config.serialize()
print(result)
self.assertDictEqual(
result,
{
"domain": "image_recognition",
"device": "cpu",
"model": {
"name": "resnet50_v1_5",
"framework": "tensorflow",
"outputs": "softmax_tensor",
},
"quantization": {
"calibration": {
"sampling_size": 100,
"dataloader": {
"batch_size": 1,
"dataset": {
"ImageRecord": {
"root": "/path/to/calibration/dataset",
},
},
"transform": {
"ParseDecodeImagenet": None,
"ResizeCropImagenet": {
"height": 224,
"width": 224,
"mean_value": [123.68, 116.78, 103.94],
},
},
},
},
"model_wise": {
"activation": {
"algorithm": "minmax"
}
},
"approach": "post_training_static_quant",
},
"evaluation": {
"accuracy": {
"metric": {
"topk": 1
},
"dataloader": {
"batch_size": 32,
"dataset": {
"ImageRecord": {
"root": "/path/to/evaluation/dataset",
},
},
"transform": {
"ParseDecodeImagenet": None,
"ResizeCropImagenet": {
"height": 224,
"width": 224,
"mean_value": [123.68, 116.78, 103.94],
},
},
},
},
"performance": {
"warmup": 10,
"iteration": -1,
"configs": {
"cores_per_instance": 3,
"num_of_instance": 2,
},
"dataloader": {
"batch_size": 1,
"dataset": {
"ImageRecord": {
"root": "/path/to/evaluation/dataset",
},
},
"transform": {
"ParseDecodeImagenet": None,
"ResizeCropImagenet": {
"height": 224,
"width": 224,
"mean_value": [123.68, 116.78, 103.94],
},
},
},
},
},
"tuning": {
"strategy": {
"name": "basic",
},
"accuracy_criterion": {
"relative": 0.01
},
"exit_policy": {
"timeout": 0
},
"random_seed": 9527,
},
},
)
def test_config_serializer(self) -> None:
"""Test Config serializer."""
config = Config(self.predefined_config)
result = config.serialize()
self.assertDictEqual(
result,
{
"domain": "image_recognition",
"device": "cpu",
"model": {
"name": "resnet50_v1_5",
"framework": "tensorflow",
"outputs": "softmax_tensor",
},
"quantization": {
"calibration": {
"sampling_size": 100,
"dataloader": {
"batch_size": 1,
"dataset": {
"ImageRecord": {
"root": "/path/to/calibration/dataset",
},
},
"transform": {
"ResizeCropImagenet": {
"height": 224,
"width": 224,
"mean_value": [123.68, 116.78, 103.94],
},
},
},
},
"model_wise": {
"activation": {
"algorithm": "minmax"
}
},
"approach": "post_training_static_quant",
},
"evaluation": {
"accuracy": {
"metric": {
"topk": 1
},
"dataloader": {
"batch_size": 32,
"dataset": {
"ImageRecord": {
"root": "/path/to/evaluation/dataset",
},
},
"transform": {
"ResizeCropImagenet": {
"height": 224,
"width": 224,
"mean_value": [123.68, 116.78, 103.94],
},
},
},
"postprocess": {
"transform": {
"LabelShift": {
"Param1": True,
},
},
},
},
"performance": {
"warmup": 10,
"iteration": -1,
"configs": {
"cores_per_instance": 3,
"num_of_instance": 2,
},
"dataloader": {
"batch_size": 1,
"dataset": {
"ImageRecord": {
"root": "/path/to/evaluation/dataset",
},
},
"transform": {
"ResizeCropImagenet": {
"height": 224,
"width": 224,
"mean_value": [123.68, 116.78, 103.94],
},
},
},
},
},
"tuning": {
"strategy": {
"name": "basic",
},
"accuracy_criterion": {
"relative": 0.01
},
"exit_policy": {
"timeout": 0
},
"random_seed": 9527,
},
"graph_optimization": {
"precisions": "bf16,fp32",
"op_wise": {
"weight": {
"granularity": "per_channel",
"scheme": "asym",
"dtype": "bf16",
"algorithm": "kl",
},
"activation": {
"granularity": "per_tensor",
"scheme": "sym",
"dtype": "int8",
"algorithm": "minmax",
},
},
},
"pruning": {
"magnitude": {
"weights": [1, 2, 3],
"method": "per_tensor",
"init_sparsity": 0.42,
"target_sparsity": 0.1337,
"start_epoch": 13,
"end_epoch": 888,
},
"start_epoch": 1,
"end_epoch": 2,
"frequency": 3,
"init_sparsity": 0.4,
"target_sparsity": 0.5,
},
},
)