Banana Launches Carrot: the GPT3 of Computer Vision

Banana Launches Carrot: the GPT3 of Computer Vision

Tiled background image of orange carrots.

We did a thing today. Banana is excited to announce the launch of a new state-of-the-art computer vision model, named Carrot, developed by our (also new) open-research community, Plantain Labs.

What is Plantain Labs?

Plantain Labs is a spin-out from Banana that is an open-research community for ML model development. Banana is singularly focused on offering the best in-class model deployment and hosting infrastructure. The development of new ML models is beyond our core mission, prompting the launch of Plantain Labs as a community-based platform devoted to model research and development.

What is Carrot?

Carrot is a cutting-edge vision-language model hosted by Banana and available as an API that is built to give application developers plug-and-play access to general-purpose image captioning, image question and answering, and image classification. People seem to think it’s pretty neat. We agree.

Carrot is a significant leap forward in the capabilities of computer vision. Previously, many use-cases required explicit training to identify a limited set of items in images and videos. Carrot can identify anything, and does so using natural English prompts - similar to GPT-3.

Sahil Chaudhary, Lead R&D Engineer behind the Carrot model at Plantain Labs says:

We saw that small, hardcoded models in natural language were replaced by the large and general purpose GPT-3 which made it accessible to build products powered by best in class AI, and Carrot goes a step beyond that, adding visual and language awareness to the same model. With so many small, use-case specific models in computer vision right now, it’s easy to imagine Carrot being the GPT-3 moment for vision.

Carrot Demo

Image Captioning:

Carrot model demo image of goldfish in water.

Generated caption:

”A large gold fish with a black shark’s head on top swimming near the surface of blue water with bubbles on it’s top, water, under the edges is gray clouds.”

Collaborative Question and Answering:

Carrot model demo image of horses and mountain.

”A group of people with their horses on the side of a hill next to a mountain covered in mounds of dirt with the top of a volcano visible into the sky above.”

Question Answer:

Q: What objects are all in the image? A: horses, people

Q: How many horses are there in the image? A: 10

Q: Is someone at the risk of falling? A: no

Is Carrot available to use for free?

Yes. Until Monday (Feb 28), Banana is offering unlimited usage of the Carrot model for FREE. After Monday, we may introduce a paid tier if you are “going bananas” with the usage of Carrot. We plan to keep a free tier available continuously after Monday.

Why offer unlimited usage of Carrot for free?

Selfishly, this is a chance to show off the scaleability and reliability of Banana’s model hosting infrastructure. But it’s also fun to share the full potential of a model like this to our community. A core value at Banana is to lower the barrier of entry to deploy and host ML models, hence why we already offer free API access to GPT-J and CLIP models.

How can I try Carrot?

Banana has a browser sandbox (rate-limited) for you to try Carrot and test out its capabilities without needing a user account. Create a free user account to get your own Carrot API key that is not rate-limited.

Banana for scale?

Banana is the easiest method to deploy, host, and serve ML models to production with scale. How easy, you ask? With Carrot we deployed and hosted the model to production using Banana in just two days.

Normally it can take weeks or months for teams to build their own deployment solution. And let’s be honest, when your team has built a mind-blowing model you want to launch it into the world ASAP. The last thing you want to do is divert weeks to build deployment infrastructure to host your model. That’s why we built Banana.

Banana helps get ML companies to software company margins. How can we make that claim?

  • Serverless GPUs = This is the biggest innovation for our customers. You only pay for the GPU resources you need (utilization time) rather than always on GPU costs which is expensive and the standard hosting method offered by other providers. We can just about guarantee you will see substantial cost savings due to this.
  • Lightning-fast latency = our product will increase your latency speeds by 5x on average. For real! Our engineering team has optimized our hosting infra to enable this speed boost.
  • Focused engineer resources = we save teams months of dev time and hiring costs to do the dirty work of building scalable hosting infra. Our sole focus is to support you in getting models into production fast and with scale.

We hope you enjoy playing with Carrot 🥕

Tag us on Twitter or join our Discord and tell us how you are using Carrot. We are giving away Banana swag to the coolest use cases and most active users!

If you are looking for an ML model hosting provider to save you time, money, and eliminate the headache of deploying to production - let’s talk! 🍌