Layer-Based Benchmark

class neurio.benchmarking.suites.LayerBasedBenchmark(seed=0, num_inference=16)

Benchmarking suite to test the inference of single layers on a given device.

The list of layer is as follows:

Block	Config-1	Config-2	Config-3	Config-4
Convolutional 1D	filters 32, stride 1	filters 32, stride 2	filters 64, stride 1	filters 64, stride 2
Convolutional 2D	filters 32, stride 1	filters 32, stride 2	filters 64, stride 1	filters 64, stride 2
Convolutional 2D	filters 32, 1x1	filters 32, 3x3	filters 32, 5x5 | filters 64, 7x7
Depthwise 2D conv	filters 32, stride 1	filters 32, stride 2	filters 64, stride 1	filters 64, stride 2
Convolutional 3D	filters 32, stride 1	filters 32, stride 2	filters 64, stride 1	filters 64, stride 2
Fully connected	128-32 neurons	256-64 neurons	512-128 neurons	1024-256 neurons
AvgPooling1D	2, stride 1	2, stride 2	4, stride 2	7, stride 1
AvgPooling2D	2x2, stride 1	2x2, stride 2	4x4, stride 2	7x7, stride 1
MaxPooling1D	2, stride 1	2, stride 2	4, stride 2	7, stride 1
MaxPooling2D	2x2, stride 1	2x2, stride 2	4x4, stride 2	7x7, stride 1
Activation	ReLU	Sigmoid	ReLU6	Tanh
Batch norm	1D	2D	3D	4D
Reshape	1D	2D	3D	4D

get_model(x, y)

Get the model at the location x,y of the LayerBasedBenchmark

Parameters:

• line_idx –

• config_idx –

Returns:

info(x=None, y=None)

Prints the structure of the LayerBasedBenchmark.

Parameters:

• line_idx – If set, prints all models in the line x from the LayerBasedBenchmark

• config_idx – If set, prints the model located at the line x and column y provided.

run_on(device: Device)

Run the Benchmark on a target device. Each model of the benchmark is compiled and deployed on the device for inference. For the inference, random data is generated according to the dimension of the input, and infered using the device infer method.

Parameters:

device – a device inheriting from Device

Returns:

the results as JSON