megenginelite.struct¶
- class LiteDeviceType(value)[source]¶
The backend device type enum
Note
compute and storage will base on the device
- LITE_ATLAS = 3¶
- LITE_CAMBRICON = 5¶
- LITE_CPU = 0¶
- LITE_CUDA = 1¶
- LITE_DEVICE_DEFAULT = 7¶
- LITE_NPU = 4¶
- class LiteDataType(value)[source]¶
The tensor data type enum
Note
half for float16, int for int32
- LITE_FLOAT = 0¶
- LITE_HALF = 1¶
- LITE_INT = 2¶
- LITE_INT16 = 3¶
- LITE_INT8 = 4¶
- LITE_UINT16 = 6¶
- LITE_UINT8 = 5¶
- class LiteTensorPhase(value)[source]¶
The tensor type enum .. note:: LITE_IO for both LITE_INPUT and LITE_OUTPUT
- LITE_INPUT = 1¶
- LITE_IO = 0¶
- LITE_OUTPUT = 2¶
- class LiteIOType(value)[source]¶
The input and output type enum, include SHAPE and VALUE sometimes user only need the shape of the output tensor
- LITE_IO_SHAPE = 1¶
- LITE_IO_VALUE = 0¶
- class LiteAlgoSelectStrategy(value)[source]¶
Operation algorithm seletion strategy type enum, some operations have multi algorithms, different algorithm has different attribute, according to the strategy, the best algorithm will be selected.
Note
These strategies can be combined LITE_ALGO_HEURISTIC | LITE_ALGO_PROFILE means: if profile cache not valid, use heuristic instead
LITE_ALGO_HEURISTIC | LITE_ALGO_REPRODUCIBLE means: heuristic choice the reproducible algo
LITE_ALGO_PROFILE | LITE_ALGO_REPRODUCIBLE means: profile the best algorithm from the reproducible algorithms set
LITE_ALGO_PROFILE | LITE_ALGO_OPTIMIZED means: profile the best algorithm form the optimzed algorithms, thus profile will process fast
LITE_ALGO_PROFILE | LITE_ALGO_OPTIMIZED | LITE_ALGO_REPRODUCIBLE means: profile the best algorithm form the optimzed and reproducible algorithms
- LITE_ALGO_HEURISTIC = 1¶
- LITE_ALGO_OPTIMIZED = 8¶
- LITE_ALGO_PROFILE = 2¶
- LITE_ALGO_REPRODUCIBLE = 4¶