Quick and simple training for AI models
Launch your AI training tasks in the cloud, without having to worry about how the infrastructure works. AI Training enables data scientists to focus on their core business, without having to worry about orchestrating computing resources.
This solution is built on the open-source Kubernetes platform, and enables you to train your models easily — in just a few clicks, or in the command line. Save time, increase your team’s productivity, and maintain the integrity of your sensitive data.
For developers and data scientists
Develop your models with popular frameworks like Scikit-learn, XGBoost and TensorFlow. Launch training tasks on one or more CPU/GPU nodes in a few seconds. All you need to run is a single line of code or an API call.
Resource optimisation
Our solution manages usage planning for your CPU/GPU computing resources. This means you do not need to factor it into your organisation.
GPU power at the best price
AI Training offers you the best prices on the market for CPU/GPU resources. Billing per minute is transparent, which simplifies the way you manage your budget, and helps you optimise your spending.
ISO/IEC 27001, 27701 and health data hosting compliance*
Our cloud infrastructures and services are ISO/IEC 27001, 27017, 27018 and 27701 certified. Thanks to our compliance, you can host healthcare data securely.
* Coming soon
Our computing resources
 CPU |
 GPU |
|
|---|---|---|
| Description | Get guaranteed CPU resources that are adapted to your test, pre-production and even basic machine learning phases. | Deploy GPU resources for all of your AI training and deep learning use cases, which require intensive processing. |
| Name | ai1-*-cpu | ai1-*-gpu |
| Applications | Basic machine learning training, tests, pre-production | Deep learning training (NLP, Computer Vision, etc.) |
| vCPU architecture | Intel Xeon processors from 1 to 12 vCores + 2GHz |
Intel Xeon processors 13 vCores + 2.9GHz |
| RAM | 4GiB RAM | 40GiB RAM |
| GPU architecture | N/A | NVIDIA TESLA V100s RAM dedicated to GPU: 32GiB GPU performance (FP32): 16.4 TFLOPS |
| Temporary local storage | 40GiB | 750GiB |
| Object Storage | Unlimited | Unlimited |
| Public network | 500Mbit/s | 1,5Gbit/s |
| Maximum number of resources per training job or notebook | 12 CPUs | 4 GPUs |
Uses
What sets the AI Training solution apart
Simplicity to accelerate your AI projects
AI Training is a technology-neutral platform. It is based on the open-source Kubernetes platform, so you can optimise the resource usage for your training, depending on your needs. This public cloud solution is commitment-free, pay-as-you-go, and dynamically adapts to your resource usage to offer you maximum flexibility and power. AI Training drastically improves productivity for data scientists, and simplifies day-to-day work by eliminating complex engineering tasks.
Lower costs, controlled billing
With AI Training, you get transparent pricing and access to the OVHcloud Control Panel, so you can easily track the cost of training tasks. This guarantees simplicity and predictability.
A European cloud that respects your data
Artificial intelligence opens up great opportunities for adding value to data, but it also requires end-to-end governance. Choosing a European cloud provider like OVHcloud ensures that your data is protected (GDPR), without the risk of a competitive breach caused by legislations such as the US Cloud Act.
Built on open and open-source standards
AI Training is natively compatible with industry-leading applications and frameworks, such as Scikit-learn, TensorFlow, Pytorch and Jupyter Notebook. Transparency is a key value of ours, and we promise total reversibility for your training processes.
Use case
Once you have created your Public Cloud project, you can launch your training tasks on computing resources directly via the OVHcloud Control Panel. You can do this via a Jupyter notebook, an API, or in the command line.
In just a few clicks, you can deploy a Jupyter notebook with the framework you want to use (TensorFlow, PyTorch, MxNet, etc.), or you can containerise your training code.
Via the OVHcloud Control Panel
Our web interface offers many features, including a dashboard to monitor your training, the ability to launch new training tasks, manage your data sets, and much more.
Via the command line
Example: You are looking to optimise a network of neurons that will help you classify your support tickets. This is how to deploy a JupyterLab with Pytorch, and your data, in two command lines.
Get the list of resources available on the AI Training platform:./ovhai capability
gpus:
- model: Tesla-V100S
max_gpus: 4
default: true
cpus:
- max_cpus: 12
Sync your local data to your object storage./ovhai data upload GRA myBucket train.zip
Run a JupyterLab with Pytorch pre-installed and mount data in /data read/write with 2 GPUs./ovhai job run \
--gpu 2 \
--name ai-training-pytorch-short-feynman \
--volume myBucket@GRA:/data:RW \
ovhcom/ai-training-pytorch:1.6.0
Retrieve results
Automatic at the end of the job
Find out more
Please refer to our technical documentation for more details on how to use AI Training.
Features
Submitting training tasks
Define the resources required for your computing tasks, and use them for AI Training. The solution works with Docker images from public or private models. You can choose from a wide catalogue of templates to get started.
Orchestration and parallelisation
AI Training orchestrates training on our infrastructure, so you can run one or more tasks in parallel. This means your teams do not need to worry about hardware resource allocation, or GPU driver compatibility.
Compatible with TensorFlow, PyTorch, Scikit-learn, XGBoost and much more
If your AI models can leverage GPU resources in a containerised environment, AI Training is the perfect tool. No matter which application you use, our solution offers unparalleled simplicity in your GPU training tasks.
Compatibility with a large catalogue of pre-installed models
A large portfolio of public Docker images is available for free, making it even easier to get started.
Deploy images designed for deep learning (autoML on text, image or video), configured to harness the power of our GPUs.
Code via notebooks
Deploy pre-configured Jupyter notebooks for your preferred frameworks, such as TensorFlow, PyTorch, Scikit-learn, and more.
Managed from the OVHcloud Control Panel, via the API or in the command line
Depending on your skills and preference, you can launch and track your computing tasks from the web interface, via the API, or via the command line. From Python to Java, Scala, C++ with Cuda, Golang, and much more — whatever programming language you want to use, anything is possible.
Logs
With quick access to your event logs, you can easily monitor your tasks.
AI Training billing
For each training task you launch, you pay per minute for the computing resources used, depending on their lifespan and the number of GPUs allocated.
No commitment is required to use this service.
