Running Neuron Apache MXNet ResNet50 on Inferentia

Introduction:

In this tutorial we will compile and deploy ResNet50 model for Inferentia. In this tutorial we provide two main sections:

1.Compile the ResNet50 model.

2.Infer the compiled model.

Before running the following verify this Jupyter notebook is running “conda_aws_neuron_mxnet_p36” kernel. You can select the Kernel from the “Kernel -> Change Kernel” option on the top of this Jupyter notebook page. Neuron supports Python module, Symbol APIs and the C predict API. The following quick start example uses the Symbol API.

Warning

This tutorial was tested on MXNet-1.5

MXNet-1.5 entered maintenance mode and require Neuron runtime 1.0, please see : MXNet-1.5 enters maintainence mode

To setup development environment for MXNet-1.5 see installation instructions for Neuron 1.15.1 : Neuron-1.15.1 MXNet install

Compile model on Neuron

The following step will compile the resnet50 model. Compilation will take a few minutes on inf1.6xlarge. At the end of compilation, the files resnet-50_compiled-0000.params and resnet-50_compiled-symbol.json will be created in local directory.

import mxnet as mx
import numpy as np

path='http://data.mxnet.io/models/imagenet/'
mx.test_utils.download(path+'resnet/50-layers/resnet-50-0000.params')
mx.test_utils.download(path+'resnet/50-layers/resnet-50-symbol.json')
sym, args, aux = mx.model.load_checkpoint('resnet-50', 0)

# Compile for Inferentia using Neuron
inputs = { "data" : mx.nd.ones([1,3,224,224], name='data', dtype='float32') }
sym, args, aux = mx.contrib.neuron.compile(sym, args, aux, inputs)

#save compiled model
mx.model.save_checkpoint("resnet-50_compiled", 0, sym, args, aux)

!ls

Deploy on Inferentia

Using same instance to deploy the model.

import mxnet as mx
import numpy as np

path='http://data.mxnet.io/models/imagenet/'
mx.test_utils.download(path+'synset.txt')

fname = mx.test_utils.download('https://raw.githubusercontent.com/awslabs/mxnet-model-server/master/docs/images/kitten_small.jpg?raw=true')
img = mx.image.imread(fname)# convert into format (batch, RGB, width, height)
img = mx.image.imresize(img, 224, 224) # resize
img = img.transpose((2, 0, 1)) # Channel first
img = img.expand_dims(axis=0) # batchify
img = img.astype(dtype='float32')

sym, args, aux = mx.model.load_checkpoint('resnet-50_compiled', 0)
softmax = mx.nd.random_normal(shape=(1,))
args['softmax_label'] = softmax
args['data'] = img

# Inferentia context
ctx = mx.neuron()

exe = sym.bind(ctx=ctx, args=args, aux_states=aux, grad_req='null')

with open('synset.txt', 'r') as f:
     labels = [l.rstrip() for l in f]

exe.forward(data=img)
prob = exe.outputs[0].asnumpy()# print the top-5
prob = np.squeeze(prob)
a = np.argsort(prob)[::-1]
for i in a[0:5]:
     print('probability=%f, class=%s' %(prob[i], labels[i]))

# Sample output will look like below:
#probability=0.634792, class=n02123045 tabby, tabby cat
#probability=0.193601, class=n02123159 tiger cat
#probability=0.103627, class=n02124075 Egyptian cat
#probability=0.031604, class=n02127052 lynx, catamount
#probability=0.015892, class=n02129604 tiger, Panthera tigris