58 lines
2.5 KiB
Python
58 lines
2.5 KiB
Python
# model
|
|
model = Model()
|
|
sparseData = Input("sparseData", "TENSOR_FLOAT32", "{8}")
|
|
traversalOrder = Parameter("traversalOrder", "TENSOR_INT32", "{4}", [0, 1, 2, 3])
|
|
blockMap = Parameter("blockMap", "TENSOR_INT32", "{2}", [0, 1])
|
|
dimFormat = Parameter("dimFormat", "TENSOR_INT32", "{4}", [0, 1, 0, 0])
|
|
dimensions = Parameter("dimensions", "TENSOR_INT32", "{4}", [2, 2, 2, 2])
|
|
d0ArrSegments = Parameter("d0ArrSegments", "TENSOR_INT32", "{0}", [])
|
|
d0ArrIndices = Parameter("d0ArrIndices", "TENSOR_INT32", "{0}", [])
|
|
d1ArrSegments = Parameter("d1ArrSegments", "TENSOR_INT32", "{3}", [0, 2, 2])
|
|
d1ArrIndices = Parameter("d1ArrIndices", "TENSOR_INT32", "{2}", [0, 1])
|
|
d2ArrSegments = Parameter("d2ArrSegments", "TENSOR_INT32", "{0}", [])
|
|
d2ArrIndices = Parameter("d2ArrIndices", "TENSOR_INT32", "{0}", [])
|
|
d3ArrSegments = Parameter("d3ArrSegments", "TENSOR_INT32", "{0}", [])
|
|
d3ArrIndices = Parameter("d3ArrIndices", "TENSOR_INT32", "{0}", [])
|
|
denseOut = Output("denseOut", "TENSOR_FLOAT32", "{4, 4}")
|
|
model = model.Operation("DENSIFY", sparseData, traversalOrder, blockMap,
|
|
dimFormat, dimensions, d0ArrSegments, d0ArrIndices, d1ArrSegments,
|
|
d1ArrIndices, d2ArrSegments, d2ArrIndices, d3ArrSegments,
|
|
d3ArrIndices).To(denseOut)
|
|
|
|
# Example 1. Input in operand 0,
|
|
input0 = {sparseData: # input 0
|
|
[1.0, 0.0, 0.0, 4.0, 2.0, 3.0, 0.0, 0.0]}
|
|
|
|
output0 = {denseOut: # output 0
|
|
[1.0, 0.0, 2.0, 3.0, 0.0, 4.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]}
|
|
|
|
quant8_symm = DataTypeConverter().Identify({
|
|
sparseData: ("TENSOR_QUANT8_SYMM", 2.125),
|
|
denseOut: ("TENSOR_QUANT8_SYMM", 2.125)
|
|
})
|
|
|
|
quant8_asymm = DataTypeConverter().Identify({
|
|
sparseData: ("TENSOR_QUANT8_ASYMM", 4.5, 2),
|
|
denseOut: ("TENSOR_QUANT8_ASYMM", 4.5, 2)
|
|
})
|
|
|
|
quant8_asymm_signed = DataTypeConverter().Identify({
|
|
sparseData: ("TENSOR_QUANT8_ASYMM_SIGNED", 2.5, 24),
|
|
denseOut: ("TENSOR_QUANT8_ASYMM_SIGNED", 2.5, 24)
|
|
})
|
|
|
|
quant16_symm = DataTypeConverter().Identify({
|
|
sparseData: ("TENSOR_QUANT16_SYMM", 3.0),
|
|
denseOut: ("TENSOR_QUANT16_SYMM", 3.0)
|
|
})
|
|
|
|
quant16_asymm = DataTypeConverter().Identify({
|
|
sparseData: ("TENSOR_QUANT16_ASYMM", 2.3, 8),
|
|
denseOut: ("TENSOR_QUANT16_ASYMM", 2.3, 8)
|
|
})
|
|
|
|
# Instantiate an example
|
|
Example((input0, output0)).AddVariations("relaxed", "float16", "int32", quant8_symm,
|
|
quant8_asymm, quant8_asymm_signed, quant16_symm,
|
|
quant16_asymm)
|