Solving the Infrastructure Bottleneck
The AIchor platform allows the user to run Machine Learning and Reinforcement Learning workloads at scale, leveraging the underlying infrastructure in the cloud from code commits
Why AIchor
AIchor's platform objective is to take advantage of the full power of high density CPU cores and advanced GPUs to distribute machine learning workloads at scale while abstracting the hardware and technical intricacies to AI Engineers and researchers.
-
Optimize the cost of infrastructure by using a containerized environment
-
Schedule machine learning experiments at scale
-
Rely on a reliable solution with higher dataset durability
-
Have a seamless user experience and empower AI engineers
Seamless Distribution
A secure platform where Compute requirements specifications can be easily bootstrapped in a single file. Having one centralized access to the clusters.
GitOps Based
Researchers and Engineers can trigger experiments and trainings with a single command-line on their working branches. GitOps platform fosters collaboration and is efficient to Improve teamwork consolidation to address individual ML codes drifting.
Secure and easy to use
Workers are provisioned automatically based on the YAML description and the workload requirements taking into account the infrastructure availability.
How It Works
Get your workloads running on distributed resources in 4 steps
1 Create a Cluster
Deploy a GKE cluster from Aichor web interface where your workload will be running.
2 Plug your VCS to Aichor
Connect your repository (on Github, Gitlab or Bitbucket) to provide Aichor with access to your code.
3 Configure your job
Add a manifest file to your repository to specify resources (RAM, CPU, GPU) required for your run.
4 Trigger your job
Run your job with a simple commit and monitor its progress, logs and real-time resources used.
Workload Management
Logs management
Realtime logs being streamed and accessible via the UI
Resources
Users can monitor and manage their workloads from a web page, no need for ssh access. They can track compute costs easily by teams or projects from the user interface.
Storage
Ability to interact with input and output datasets.
Working with big pre-trained models is challenging in terms of hardware requirements as we are all familiar with the CUDA OUT OF MEMORY error. We found AIchor to be the perfect solution that not only fixed all of that but also optimized the cost of infrastructure by using a containerized environment and engaging the A100s only when called. Thanks to AIchor, we now only need a data scientist with docker skills in order for us to iterate and improve our models.
Request a Demo
Test AIchor now with a Free demo.