TensorPool Launches: The Easiest Way To Use GPUs 🌊

TensorPool recently launched!

‍

^{"Just describe your job and they handle all the GPU orchestration and execution at half the cost"}

^{‍
TL;DR: TensorPool's CLI makes ML model training effortless - just describe your job, and they handle GPU orchestration and execution at half the cost of major cloud providers.}

‍

‍Founded by Tycho Svoboda, Joshua Martinez & Hlumelo Notshe‍‍

Joshua, Tycho, and Hlumelo all met their freshman year at Stanford where they got placed on the same dorm floor. Since then, they’ve taken all their classes together and have worked at companies like Apple, Blackstone, DeepMind, Nextdoor, CZI, and PIMCO.

‍

Get setup in 3 minutes

Here’s a quick demo of how easy it is to train ML models with TensorPool!

TensorPool Demo -- Fine-Tuning DeepSeek

‍Problem

Since high school, when they trained their first ML models, accessing remote compute has always been a massive hurdle. As the years went by and they moved further into academia/industry, a good solution never came. Whether it was AWS EC2’s absurd amount of configuration, or Google Colab’s inability to run jobs without your computer being on, nothing left them satisfied. They talked to their co-workers and friends and they all felt the same. They wanted using GPUs to be as easy as programming locally.

‍

Solution

The team at TensorPool got together and built a way to access industry-scale compute from your IDE. No more SSHing. No more massive data migrations. No more ML Ops headache. Just a simple CLI. [They also cut your costs by 50% which will be explained below :)]

‍

How It Works

TensorPool has 2 main advantages:

1. Ease of Use:
   • Don’t leave your IDE: TensorPool deploys your code directly to GPUs and ships the results back, just as if you were developing locally.
   • Intuitive CLI & Configuration: With TensorPool you can run your job with a single command, kick off several experiments with different tp.config.toml configurations, and version control training jobs with your code.
2. Price Advantage
   • Multi-cloud integration: They have developed technology to do a real time analysis of all available GPU cloud providers, and put your job on the cheapest one.
   • Spot node resummation tech: They have created a way to give users the price advantage of spot nodes with the reliability of on-demand instances. This leads to ~50% savings!

‍

Try TensorPool today!

Below is a quick start:

After completing the prerequisites, they create a “tp.config.toml” (explained here), either manually with:

‍

Or auto-generate one with natural language:

‍

To then deploy and run your job – simply run:

‍

and that’s it! Your model is training on cloud GPUs!

‍

Learn More

^‍

^{🌐 Visit tensorpool.dev to learn more.}

^‍

^{💥 They are currently giving away $5/week of free compute to early users, and would love for you all to try them out! Whether your start-up needs an ML infra upgrade or you love messing around with models in your free time – reach out to them, they would love to chat!}

^‍

^{🤝 They would also love intros to enterprises that you believe could benefit from cheaper / easier access to GPUs.}

<sup>‍

🌟 Give TensorPool a star on Github & get set up in 3 minutes!</sup>

^‍

^{‍‍👣 Follow TensorPool on LinkedIn & X.}