Compile Options
- class Framework(value)[source]
Bases:
str
,Enum
An enumeration.
- PYTORCH = 'pytorch'
- ONNX = 'onnx'
- ONNXRUNTIME = 'onnxruntime'
- AIMET = 'aimet'
- TENSORFLOW = 'tensorflow'
- TFLITE = 'tensorflow_lite'
- COREML = 'coreml'
- TENSORRT = 'tensorrt'
- QNN = 'qnn'
- class Extension(value)[source]
Bases:
str
,Enum
An enumeration.
- ONNX = '.onnx'
- PT = '.pt'
- AIMET = '.aimet'
- H5 = '.h5'
- class Runtime(value)[source]
Bases:
str
,Enum
An enumeration.
- TFLITE = 'tflite'
- QNN_LIB_AARCH64_ANDROID = 'qnn_lib_aarch64_android'
- QNN_CONTEXT_BINARY = 'qnn_context_binary'
- ONNX = 'onnx'
- PRECOMPILED_QNN_ONNX = 'precompiled_qnn_onnx'
- class QuantizeFullType(value)[source]
Bases:
str
,Enum
An enumeration.
- INT8 = 'int8'
- INT16 = 'int16'
- W8A16 = 'w8a16'
- W4A8 = 'w4a8'
- W4A16 = 'w4a16'
- class CompileOptions(compute_unit: List[ComputeUnit] | None = None, target_runtime: Runtime | None = Runtime.TFLITE, output_names: str | None = None, truncate_64bit_tensors: bool | None = False, truncate_64bit_io: bool | None = False, force_channel_last_input: str | None = None, force_channel_last_output: str | None = None, quantize_full_type: QuantizeFullType | None = None, quantize_weight_type: QuantizeWeightType | None = None, quantize_io: bool | None = False, quantize_io_type: str | None = None, qnn_graph_name: str | None = None, qnn_context_binary_vtcm: str | None = None, qnn_context_binary_optimization_level: int | None = None)[source]
Bases:
CommonOptions
Compile options for the model.
Note
For details, see CompileOptions in QAI Hub API.
- output_names: str | None = None
- truncate_64bit_tensors: bool | None = False
- truncate_64bit_io: bool | None = False
- force_channel_last_input: str | None = None
- force_channel_last_output: str | None = None
- quantize_full_type: QuantizeFullType | None = None
- quantize_weight_type: QuantizeWeightType | None = None
- quantize_io: bool | None = False
- quantize_io_type: str | None = None
- qnn_graph_name: str | None = None
- qnn_context_binary_vtcm: str | None = None
- qnn_context_binary_optimization_level: int | None = None