azure-samples / aml-hardware-accelerated-models Goto Github PK

View Code? Open in Web Editor NEW

6.0 19.0 9.0 5.42 MB

Sample client code for Azure ML HW Accelerated Models Powered by Project Brainwave

License: MIT License

Dockerfile 0.34% Jupyter Notebook 18.91% Python 72.23% C# 8.52%

aml-hardware-accelerated-models's Introduction

page_type

languages

products

description

urlFragment

sample

python

csharp

azure

These samples assume you are familiar with the Azure ML Hardware Accelerated Models (HAM) product.

aml-hardware-accelerated-models

Azure Machine Learning Hardware Accelerated Models - Samples

Getting Started

These samples assume you are familiar with the Azure ML Hardware Accelerated Models (HAM) product. If not, you can read more here. The samples in this repo are:

Deploy to Data Box Edge
- Follow these sample notebooks if you have a Databox Edge machine. Data Box Edge is an on-premise server that is enabled with an FPGA. Read more here.

Quickstart

git clone https://github.com/Azure-Samples/aml-hardware-accelerated-models.git
cd aml-hardware-accelerated-models
pip install azureml-accel-models[cpu]
pip install jupyter
jupyter notebook

aml-hardware-accelerated-models's People

Contributors

Stargazers

Watchers

Forkers

borisneal holzman shrib stjordanis millertracy flamingofugang buswrecker garvitarai isabella232

aml-hardware-accelerated-models's Issues

Deploy to Edge via SDK - missing pip install azureml-contrib-iot

ipynb fails at Line 18

from azureml.contrib.core.compute import IotHubCompute


# If you haven't attached your Data Box Edge's IoT Hub as compute
config = IotHubCompute.attach_configuration(name=iot_hub_name, resource_group=resource_group, connection_string=connection_string)
iothub_compute = IotHubCompute.attach(ws, iot_device_id, config)
iothub_compute.wait_for_completion()



---------------------------------------------------------------------------
ImportError                               Traceback (most recent call last)
<ipython-input-18-9759e54235ca> in <module>
      1 #from azureml.contrib.iot.compute import IotHubCompute
----> 2 from azureml.contrib import IotHubCompute
      3 
      4 # If you haven't attached your Data Box Edge's IoT Hub as compute
      5 config = IotHubCompute.attach_configuration(name=iot_hub_name, resource_group=resource_group, connection_string=connection_string)

ImportError: cannot import name 'IotHubCompute'

Remedy is to edit package list

Change this code block

%%bash
pip install azure-cli # Install Azure CLI
az extension add --name azure-cli-iot-ext # Add IoT CLI extension
pip install -U jupyter_console # Fix Jupyter dependencies overridden by Azure CLI installation
pip install docker # Install Docker
pip install azureml-accel-models[cpu]
az login # Log into Azure account

and include

pip install azureml-contrib-iot

support for tensorflow==2.x.x

Client App crashes at Line 56 of run.py

if incorrect version of tensorflow is used; an error occurs
Specifically at

# Using the grpc client in AzureML Accelerated Models SDK
from azureml.accel import PredictionClient
prediction_client = PredictionClient(args.address, args.port)

results = prediction_client.score_file(path=os.path.join(args.image_dir, image), 
                                                        input_name=args.input_tensors, 
                                                        outputs=args.output_tensors

Error Message from IOTHub at Databox Edge

Trying again in 10 seconds...
IoT Hub responded to message with status: OK
Unable to inference for unknown reason. See stack trace below:
Traceback (most recent call last):
  File "run.py", line 56, in main
    outputs=args.output_tensors)
  File "/usr/local/lib/python3.5/dist-packages/azureml/accel/_client.py", line 77, in score_file
    result = self.score_tensors(input_map, outputs, timeout)
  File "/usr/local/lib/python3.5/dist-packages/azureml/accel/_client.py", line 128, in score_tensors
    return self.__predict(request, timeout, outputs=outputs)
  File "/usr/local/lib/python3.5/dist-packages/azureml/accel/_client.py", line 160, in __predict
    return tf.contrib.util.make_ndarray(result.outputs[output_names])
AttributeError: module 'tensorflow' has no attribute 'contrib'

pip list | grep azureml

azureml-accel-models                  1.0.76       
azureml-automl-core                   1.0.76       
azureml-automl-runtime                1.0.76.1     
azureml-contrib-datadrift             1.0.74       
azureml-contrib-interpret             1.0.74       
azureml-contrib-iot                   1.0.76       
azureml-contrib-notebook              1.0.76       
azureml-contrib-opendatasets          1.0.45       
azureml-contrib-pipeline-steps        1.0.76       
azureml-contrib-reinforcementlearning 0.1.0.5919674
azureml-contrib-server                1.0.76       
azureml-contrib-services              1.0.76       
azureml-core                          1.0.76       
azureml-datadrift                     1.0.74       
azureml-dataprep                      1.1.30       
azureml-dataprep-native               13.1.0       
azureml-defaults                      1.0.76       
azureml-explain-model                 1.0.74       
azureml-interpret                     1.0.74       
azureml-mlflow                        1.0.74       
azureml-model-management-sdk          1.0.1b6.post1
azureml-opendatasets                  1.0.74       
azureml-pipeline                      1.0.76       
azureml-pipeline-core                 1.0.76       
azureml-pipeline-steps                1.0.76       
azureml-samples                       0+unknown    
azureml-sdk                           1.0.76       
azureml-telemetry                     1.0.76       
azureml-tensorboard                   1.0.74       
azureml-train                         1.0.76       
azureml-train-automl                  1.0.76       
azureml-train-automl-client           1.0.76       
azureml-train-automl-runtime          1.0.76.1     
azureml-train-core                    1.0.76       
azureml-train-restclients-hyperdrive  1.0.76       
azureml-widgets                       1.0.76

Fix can be done by changing Dockerfile

RUN pip3 install --upgrade pip && \
    pip install setuptools && \
    pip install tensorflow==1.5.0

Inference fails as a script, but not interactive

Please provide us with the following information:

This issue is for a: (mark with an `x`)

- [X ] bug report -> please search issues before submitting
- [ ] feature request
- [ ] documentation issue or request
- [ ] regression (a behavior that used to work and stopped in a new release)

Minimal steps to reproduce

docker run -i -t ubuntu:18.04 bash
apt update && apt-get install -y libgtk2.0-dev python3.6 python3-pip openssh-client vim
pip3 install opencv-python azureml-accel-models tensorflow==1.6 #1.14 did not work

copy the content from https://github.com/Azure-Samples/aml-hardware-accelerated-models/tree/master/finetune-ssd-vgg/tfssd to /tfssd

i use scp -r [email protected]:/home/kevinsay/test/tfssd/* /tfssd/

after running https://github.com/Azure-Samples/aml-hardware-accelerated-models/blob/master/finetune-ssd-vgg/notebooks/Finetune%20VGG%20SSD.ipynb, copy the content from ~/azml_ssd_vgg/vggssd/1.1.3 to /model

i use scp -r [email protected]:/home/kevinsay/test/model/* /model/

Run python3 and past the following into the interactive shell and it works fine

Save the following as test.py and attempt to run python3 test.py == you will never see the line "INFO:tensorflow:Restoring parameters from /model/vggssd/1.1.3/ssd_vgg_bw"

import sys
import cv2
import os
import tensorflow as tf
sys.path.insert(0, '/tfssd/')
from finetune.inference import InferVggSsd
checkpointDirectory = "/model"
useGPU = False
infer = InferVggSsd(checkpointDirectory, gpu=useGPU)
print("done")

Any log messages given by the failure

None

Expected/desired behavior

I expected to see "INFO:tensorflow:Restoring parameters from /model/vggssd/1.1.3/ssd_vgg_bw" when running in script and I did not.

OS and Version?

Windows 7, 8 or 10. Linux (which distribution). macOS (Yosemite? El Capitan? Sierra?)
Ubuntu 18.04

Versions

Mention any other details that might be useful

Thanks! We'll be in touch soon.