You’ve got questions?
We’ve got answers!


We’re working in <insert name of cloud>. Can I still use SwarmOne for AI Training?

Yes. You will continue to use the cloud of your choice for any service you desire. AI Training on SwarmOne does not require you to change any aspect of your workflow. SwarmOne is specifically focused on AI training.

Can I use data in an S3 Bucket?

Yes. When you train in the Cloud you have to transfer your data every time you train. (ugh) When you train on SwarmOne you only need to migrate your data one time, and even that is done automatically. We call this super-feature YOTO (You only Transfer Once!)


I have a large dataset. Are there any limitation on the size of my data set?

No. There are no limits to the size of your datasets.

Are there any limits on training batch size?

No. There are no limits on batch size.

Can I train an LLM?

Yes. We have existing customers working with 100’s GBs of data in their data sets and tens of billions of parameters.


What Training tools do you support?

We support PyTorch (with Lightning), HuggingFace and TensorFlow

Can I work locally?

Yes. You’ll continue to work in your own notebook.

What if we usually train “on prem” in our own Data Center?

It’s easy to use both – training on your own servers when GPUs are available.We think you’ll find that training on the Swarm will be faster and easier than any other option.

Can I use GIT?

Yes. Jupyter Lab has a Git extension.You can also work with Git directly from your private jupyter server terminal.

Can I integrate SwarmONE into a production pipeline?

Yes. It’s easy to replace your current production pipeline’s training function with SwarmOne.

Can I run experiments in parallel?

Yes. The Swarm is designed to process tasks in parallel for greater speed. You can send an unlimited number of tasks.

Where does the actual training occur?

Training runs on multiple GPUs. Those GPUS can be at any one of our certified data centers.

Can I use SwarmOne for inference?

At present we are focussed on training. Inference is in our product road map but is not currently supported.


How much do most people save with SwarmOne?

In terms of “hard costs” our platform costs 33% LESS than AWS. However, the real savings come from dramatically reduced effort. Data Scientists work far more efficiently. They LOVE the platform. All of the supporting team members (MLOps, SecOps, FinOps) can go do other things.


What tools do you integrate with?

SwarmOne currently integrates with:

● Weights and Biases

● TensorBoard

● Hugging Face

● ML Flow

● DagsHub