mkl-dnn | Fork of oneAPI Deep Neural Network Library for Intel® Open | Machine Learning library

I am writing an application which requires very low latency. The application will be running on intel Xenon processor enabled with mkl-dnn instructions/AVX instructions set. The following code is taking 22 milliseconds when executed on intel 9750H processor.

ENVIRONMENT

• followed guide - https://github.com/rapidsai-community/notebooks-contrib/blob/branch-0.14/intermediate_notebooks/E2E/synthetic_3D/rapids_ml_workflow_demo.ipynb
• conda create -n rapids-0.16 -c rapidsai -c nvidia -c conda-forge -c defaults rapids=0.16 python=3.7 cudatoolkit=10.2
• AWS EC2: Deep Learning AMI (Ubuntu 18.04) Version 36.0 - ami-063585f0e06d22308: MXNet-1.7.0, TensorFlow-2.3.1, 2.1.0 & 1.15.3, PyTorch-1.4.0 & 1.7.0, Neuron, & others. NVIDIA CUDA, cuDNN, NCCL, Intel MKL-DNN, Docker, NVIDIA-Docker & EFA support. For fully managed experience, check: https://aws.amazon.com/sagemaker
• AWS EC2 instance - g4dn.4xlarge - 16GB VRAM, 64 GB RAM

CODE

• I am just trying to gave a trainign and a test set for the model
• 1st data package - train_data = xgboost.DMatrix(data=X_train, label=y_train) Up until I run just this and do training and anything with, only this does not gives an error message
• 2nd data package - test_data = xgboost.DMatrix(data=X_test, label=y_test) couple cells down the line, they are not executed together

Side Note

• ERROR GB VRAM sizes are NOT 30GB or 15GB
  • 1 539 047 424 = 1.5 GB,
  • 3 091 258 960 = 3 GB,
  • 3 015 442 432 = 3GB,
  • 3 091 258 960 = 3 GB.
  • The GPU has 16 GB VRAM, so I don't think that this answers the question.

ERROR

I am trying to create a simple Jupyter notebook in Google Cloud Platform with GPU:

• Name: PyTorch
• Region: us-west1 (Oregon)
• Zone: us-west1-b
• Operating System" Debain 9
• Environment: PyTorch 1.4 (with Intel (R) MKL-DNN/MKL)
• Machine type: n1-standard-4 (4vCPUs, 15GB RAM)
• GPU type: NVIDIA Tesla K80
• Number of GPUs: 1
• Install NVIDIA GPU driver automatically for me
• Boot disk type: Standard Persistent Disk
• Boot disk size in GB: 100
• Data disk type: Standard Persistent Disk
• Data disk size in GB: 100
• Google-managed-key
• Network: default
• Subnetwork: default
• External IP (Automatic)
• Access to JupyterLab (Service account)
• Use Compute Engine default service account

After I press "Create" it returns to a list of instances. It shows loading and after I refresh disappears. When I create a Jupiter notebook without GPU it succeeds. My guess is that I need to request a GPU quota. For this, I go to Quotas page and see that Compute Engine API for NVIDIA K80 GPUs is "1 of 24 quotas are reaching limit". When I press all quotes, I see that Current Usage is 0 for everyone 7-day peak usage is 1 fo for us-west1 and Limit is 1. I can not select any checkbox.

How can I resolve this problem? Thank you!

I am currently self-teaching myself tensorflow and I am trying to this code right here

I'm trying to build Pytorch on windows using visual studio, but it seems it faces some internal compiler error which I have not been able to figure out its cause. out of 46 targets, 35 gets built successfully until it ultimately fails with the following errors. Before I list the errors this is how I went about building it :

I am trying to programatically create "AI Platform Notebooks" in GCP. The gcloud sdk does have some support for managing these notebooks, but not creating them. And there is no client library for support for Node.js (the language I'm using). Creating notebooks is however supported by the GCP REST API as documented here. However, I am struggling to work out how to specify what notebook I want in the request's JSON. From the GCP web UI, the settings I want are:

• instance name: "testing-instance"
• region: "europe-west2"
• zone: "europe-west2a"
• environment: "TensorFlow Enterprise 2.1 (with Intel® MKL-DNN/MKL)"
• machine type: "e2-highmem-2 (Efficient Instance, 2 vCPUs, 16 GB RAM)"
• access to jupyter lab: "Single user only"
• user email: "firstname.surname@email.com"
• service account: "team@project.iam.gserviceaccount.com"

But I am struggling to translate this into a JSON request for the REST API. Below is what I have so far. I am not sure any of it is correct, and I'm definitely missing the environment (tensorflow 2.1) and single user only access. I have no idea how to go about achieving this other than randomly trying different requests until it works. (I have left some of the JSON as just specifying the types as per the docs for now, for reference).

I am trying to run the Detectron2 module on Colab using CUDA version 10.0 but since today there have been some issues regarding the versions of Cuda Compiler.

The output I get after running !nvidia-smi is :

After creating a new notebook instance in the last few days there is an internal error relating to the Git server extension when opened:

Internal Error:

Fail to get the server root path. Error: Git server extension is unavailable. Please ensure you have installed the JupyterLab Git server extension by running: pip install --upgrade jupyterlab-git. To confirm that the server extension is installed, run: jupyter serverextension list.

This means I can't use the Git clone button which returns:

Clone failed:

JSON.parse: unexpected character at line 1 column 1 of the JSON data

Has this been occurring for others? I tried the pip install --upgrade jupyterlab-git as suggested but that didn't seem to fix anything. Below is my notebook instance setup (if not specified it is the default):

Region: us-west1 (Oregon)

Zone: us-west1-a

Environment: TensorFlow Enterprise 2.1 (with Intel® MKL-DNN/MKL and CUDA 10.1)

Machine type: n1-standard-4 (4vCPUs, 15 GB RAM)

GPU type: NVIDIA Tesla T4

Number of GPUs: 1

✓ Install NVIDA GPU driver automatically for me

I'm still quite new with both Google Cloud and this is my first Stackoverflow post.

OpenVINO has some 30MB libmkl_tiny_tbb.so, which is "The special version of MKL dynamic library packed specially to use within Inference Engine library.", as stated in version.txt.

MKL-DNN has 125MB libmklml_gnu.so. Is there a way to build a ~30MB file from MKL-DNN git hub repo?