https://github.com/onnx/onnx/blob/master/docs/ImplementingAnOnnxBackend.md SingaRep provides the intermediate representation of Singa, the user can run the forward of the singa model by run func, or, the user can append more layers after the singa_ops to do t
(self, params, inputs, outputs, layers, device)
| 1951 | class SingaRep(BackendRep): |
| 1952 | |
| 1953 | def __init__(self, params, inputs, outputs, layers, device): |
| 1954 | """ |
| 1955 | https://github.com/onnx/onnx/blob/master/docs/ImplementingAnOnnxBackend.md |
| 1956 | SingaRep provides the intermediate representation of Singa, |
| 1957 | the user can run the forward of the singa model by run func, |
| 1958 | or, the user can append more layers after the singa_ops to do |
| 1959 | the transfer learning |
| 1960 | Args: |
| 1961 | params (dict{}): a dict of params, data type is numpy ndarray |
| 1962 | inputs (ValueInfo): a dict of inputs |
| 1963 | outputs (ValueInfo): a dict of outputs |
| 1964 | layers (namedtuple('operator_tuple', ['node', 'operator'])[]): a list of singa operator |
| 1965 | device (string): CPU or CUDA |
| 1966 | """ |
| 1967 | super(SingaRep, self).__init__() |
| 1968 | self.inputs = inputs |
| 1969 | self.states = params |
| 1970 | self.outputs = outputs |
| 1971 | self.dev = cpu_dev if device == "CPU" else gpu_dev |
| 1972 | self.layers = layers |
| 1973 | self.tensor_count = {} |
| 1974 | self.has_initialized = False |
| 1975 | self.is_graph = False |
| 1976 | |
| 1977 | def initialize(self): |
| 1978 | """ |