Developers looking to take their AI compute to the next level can now access the power of Graphcore IPUs on-demand, with the launch of premium notebooks tiers for Paperspace Gradient.
The new pay-as-you-go options allow commercial users and AI researchers to quickly and affordably scale-up their IPU capability to meet their growing compute requirements.
Paperspace Gradient is a powerful alternative to Google Colab - offering browser-based AI notebooks running on Graphcore’s state-of-the-art Intelligence Processing Units.
Developers have access to a wide range of IPU-optimised models, spanning NLP, Computer Vision and Graph Neural Networks. Hugging Face users can also train and fine-tune sophisticated transformer models with only a few lines of code.
For those just starting out on their IPU journey, Paperspace also offers six hours of free compute time, allowing users to try out Gradient Notebooks and discover the IPU advantage for themselves.
To coincide with the launch of the new premium notebook tiers, Graphcore and Paperspace are introducing Stable Diffusion for the first time on IPUs.
The wildly-popular generative AI model has received widespread acclaim for its stunning images and flexibility – offering text-to-image, image-to-image, and text-guided-in-painting.
Users can start generating images in a matter of minutes by simply launching a web-based Paperspace notebook and running the pre-trained Hugging Face Stable Diffusion model.
To learn more about running Stable Diffusion for IPUs on Paperspace Gradient notebooks, see our dedicated blog, which includes a getting started guide.
Graph Neural Networks
Graphcore IPUs are increasingly being recognised for their standout ability to run Graph Neural Networks.
Graphcore-developed GNNs landed two first place positions in this year’s Open Graph Network Large Scale Challenge, the AI industry’s leading test of graph AI performance.
The two winning models – GPS++ for molecular prediction and Distributed KGE TransE (256) for knowledge graph completion - are among a range of GNN models available to run (training and inference) on Paperspace Gradient Notebooks.
Other new IPU-powered Notebooks on Paperspace
We are constantly adding new notebooks for IPUs, spanning a wide range of ML solutions and applications, to Paperspace Gradient.
Among the latest additions are:
Natural Language Processing
A full list of available Paperspace Gradient Notebooks for Graphcore IPUs can be found in the table below or here.
The launch of paid notebook tiers on Paperspace reflects the growing number of users looking to scale up their IPU compute resource for commercial use. We share the belief that AI innovators want to focus on the use of artificial intelligence to make a difference, without getting bogged-down in computational complexity and time-sapping systems management.
As part of our joint commitment to growing AI businesses, we will soon be introducing a suite of new tools, designed to help users automate the process of training and deploying models for real-world use.
Gradient Workflows allows developers to automatically update their models whenever they push new code to a linked GitHub repo. The Gradient model repository gives users all the tools they need to import and manage models, as well as looking after version control.
Deployments turns trained models into API endpoints, helping customers build products and services with their same speed and ease-of-use that they have come to expect from Paperspace.
More about Graphcore and Paperspace
Users of Graphcore IPUs on Paperspace can take advantage of a wide range of resources to help them get started, run proof-of-concept models, fine-tune for their own specific needs and to scale-up their deployment.
Below is a selection of useful links:
Tutorial: Get up and running with Graphcore IPUs on Paperspace
Tutorial: Rapidly prototyping NLP apps using Hugging Face and IPUs on Paperspace
Blog: Getting started with Stable Diffusion for IPUs on Paperspace
Be part of the conversation: join the Graphcore community on Slack
IPUs on Paperspace Gradient: pricing