afe.apis.prerelease_v1

Source: afe/apis/prerelease_v1.py

Back to index

This is the pre-release API for AFE. It supports importing models, loading and storing AFE's internal format, quantizing, executing, and simulating.

Imports

• afe.apis.defines.gen1_target
• afe.ir.defines.Status
• afe.ir.tensor_type.ScalarType
• csv
• numpy as np
• os
• sima_utils.common.Platform
• sima_utils.data.data_generator.DataGenerator
• subprocess
• tempfile
• typing.Dict
• typing.List
• typing.Optional
• typing.Tuple

Functions

• load_onnx_model(model_path: str, shape_dict: Dict[str, Tuple[int, ...]], dtype_dict: Dict[str, ScalarType], layout: str = 'NCHW', is_quantized: bool = False) -> _AwesomeNet (line 42): Loads an ONNX model into an _AwesomeNet format.

  Parameters:

  • model_path: Path to a .onnx file containing the onnx model
  • shape_dict: Dictionary of input names to input shapes eg. {'input', (1,224,224,3)}
  • dtype_dict: Dictionary of input names to input types eg. {'input', 'float32'}
  • layout: Input data layout. Default is 'NCHW'
  • is_quantized: Whether the model is pre-quantized. Default is False

  Returns: An _AwesomeNet

• load_pytorch_model(model_path: str, input_names: List[str], input_shapes: List[Tuple[int, ...]], input_dtypes: Optional[List[ScalarType]] = None, layout: str = 'NCHW', is_quantized: bool = False) -> _AwesomeNet (line 62): Loads a PyTorch model into an _AwesomeNet format.

  Parameters:

  • model_path: Path to a PyTorch file (.pt) that contains the entire model
  • input_names: List of input names. eg ['input']
  • input_shapes: List of input shapes corresponding to input names. eg. [(1, 224, 224, 3)]
  • input_dtypes: List of input datatypes corresponding to input names. eg ['float32']
  • layout: Input data layout. Default is 'NCHW'
  • is_quantized: Whether the model is pre-quantized. Default is False

  Returns: An _AwesomeNet

• load_tensorflow_model(model_path: str, shape_dict: Dict[str, Tuple[int, ...]], output_names: List[str], layout: str = 'NHWC', is_quantized: bool = False) -> _AwesomeNet (line 84): Loads a TensorFlow model into an _AwesomeNet format

  Parameters:

  • model_path: Path to a .pb TensorFlow model
  • shape_dict: Dictionary of input names to input shapes eg. {'input', (1,224,224,3)}
  • output_names: List of output names of the network eg. ['output1', 'output2']
  • layout: any variation of the characters NHWC representing Batch Size, Height, Width, and Channels
  • is_quantized: Whether the model is pre-quantized. Default is False

  Returns: An _AwesomeNet

• load_tensorflow2_model(model_path: str, shape_dict: Dict[str, Tuple[int, ...]], output_names: List[str], layout: str = 'NHWC', is_quantized: bool = False) -> _AwesomeNet (line 104): Loads a TensorFlow model into an _AwesomeNet format

  Parameters:

  • model_path: Path to a SavedModel tensorflow model.
  • shape_dict: Dictionary of input names to input shapes eg. {'input', (1,224,224,3)}
  • output_names: List of output names of the network eg. ['output1', 'output2']
  • layout: any variation of the characters NHWC representing Batch Size, Height, Width, and Channels
  • is_quantized: Whether the model is pre-quantized. Default is False

  Returns: An _AwesomeNet

• load_tflite_model(model_path: str, shape_dict: Dict[str, Tuple[int, ...]], dtype_dict: Dict[str, ScalarType], layout: str = 'NHWC', is_quantized: bool = False) -> _AwesomeNet (line 124): Loads a TensorFlow Lite model into an _AwesomeNet format.

  Parameters:

  • model_path: Path to a .tflite file containing the TensorFlow Lite model
  • shape_dict: Dictionary of input names to input shapes eg. {'input', (1,224,224,3)}
  • dtype_dict: Dictionary of input names to input types eg. {'input', 'float32'}
  • layout: Input data layout. Default is 'NHWC'
  • is_quantized: Whether the model is pre-quantized. Default is False

  Returns: An _AwesomeNet

• load_keras_model(model_path: str, shape_dict: Dict[str, Tuple[int, ...]], layout: str = 'NCHW', is_quantized: bool = False) -> _AwesomeNet (line 144): Loads a Keras model into an _AwesomeNet format.

  Parameters:

  • model_path: Path to a .h5 file containing the keras model
  • shape_dict: Dictionary of input names to input shapes eg. {'input', (1,224,224,3)}
  • layout: Input data layout. Default is 'NCHW'
  • is_quantized: Whether the model is pre-quantized. Default is False

  Returns: An _AwesomeNet

• load_caffe_model(prototxt_file_path: str, caffemodel_file_path: str, shape_dict: Dict[str, Tuple[int, ...]], dtype_dict: Dict[str, ScalarType], layout: str = 'NCHW', is_quantized: bool = False) -> _AwesomeNet (line 162): Loads a caffe model into an _AwesomeNet format.

  Parameters:

  • prototxt_file_path: filepath to the caffe .prototxt file
  • caffemodel_file_path: filepath to the caffe .caffemodel file
  • shape_dict: Dictionary of input names to input shapes eg. {'input', (1,224,224,3)}
  • dtype_dict: Dictionary of input names to input types eg. {'input', 'float32'}
  • layout: Input data layout. Default is 'NCHW'
  • is_quantized: Whether the model is pre-quantized. Default is False

  Returns: An _AwesomeNet

• load_caffe2_model(init_net_file_path: str, predict_net_file_path: str, shape_dict: Dict[str, Tuple[int, ...]], dtype_dict: Dict[str, ScalarType], layout: str = 'NCHW', is_quantized: bool = False) -> _AwesomeNet (line 184): Loads a caffe2 model into an _AwesomeNet format.

  Parameters:

  • init_net_file_path: filepath to the caffe2 .pb init_net file
  • predict_net_file_path: filepath to the caffe2 .pb predict_net file
  • shape_dict: Dictionary of input names to input shapes eg. {'input', (1,224,224,3)}
  • dtype_dict: Dictionary of input names to input types eg. {'input', 'float32'}
  • layout: Input data layout. Default is 'NCHW'
  • is_quantized: Whether the model is pre-quantized. Default is False

  Returns: An _AwesomeNet

• quantize_net(net: _AwesomeNet, input_generator: DataGenerator, asymmetry: bool, per_channel: bool, max_calibration_samples: int) (line 209): Quantizes a network.

  Parameters:

  • net: an _AwesomeNet
  • input_generator: a DataGenerator we use to feed inputs into the _AwesomeNet during calibration
  • asymmetry: If True this function performs asymmetric quantization. Otherwise, it performs symmetric quantization
  • per_channel: If True this function performs per_channel quantization.
  • max_calibration_samples: Maximum number of samples we use for calibration

• execute_net(net: _AwesomeNet, inputs: Dict[str, np.ndarray], dequantize_outputs: bool = True) -> List[np.ndarray] (line 238): Executes a network.

  Parameters:

  • net: an _AwesomeNet
  • inputs: Dictionary of placeholder node names (str) to the input data
  • dequantize_outputs: Whether to dequantize the output data to floating point :return The result of executing the output node

  Returns: List[np.ndarray]

• compile_net(net: _AwesomeNet, output_elf_path: str, compress: bool = True) (line 259): Compile a network using Product Compiler.

  Parameters:

  • net: an _AwesomeNet.
  • output_elf_path: Path in which elf files should be created.
  • compress: If True mlc file is compressed before generating .elf file.

• simulate_isim_module(elf_file: str, output_path: str, output_kpi_file_name: Optional[str] = None, output_trace_file_name: Optional[str] = None, generate_csv_kpi: Optional[bool] = True, override_existing: Optional[bool] = False, target: Optional[Platform] = None) (line 277): Wrapper to take elf file and generate summary KPI file, and trace file.

  Parameters:

  • elf_file: Path to the location of the elf file to simulate in ISIM.
  • output_path: Location of where to store all output files. If location does not exist, it will be created.
  • output_kpi_file_name: Name for output kpi file. Default: elf_file name without '.elf' extension and finishing with '_kpis.txt' postfix.
  • output_trace_file_name: Name for output trace file. Default: elf_file name without '.elf' extension and finishing with '_trace.mpack' postfix.
  • generate_csv_kpi: Whether to generate a CSV file with the KPIs in addition to the txt.
  • override_existing: Whether to override existing kpis and trace files from a previous run.
  • target: [Optional] platform target.

• isim_net(elf_file: str, output_kpi_path: str, output_trace_path: Optional[str] = None, output_kpi_csv_path: Optional[str] = None, csv_format: Optional[bool] = False, target: Optional[Platform] = None) (line 327): Take elf file and generate summary KPI file, and trace file.

  Parameters:

  • elf_file: a elf_file
  • output_kpi_path: Path in which kpi files should be created
  • output_trace_path: Optional Path in which trace files should be created. Only mla-isim supports --mpack.
  • output_kpi_csv_path: Path in which kpi files in csv format should be created
  • csv_format: Generate the Summary KPI in CVS format
  • target: [Optional] platform target.