About the role:

We're training state of the art generative models in a completely new modality where the standard of quality is significantly higher than those that came before (hands with three fingers is fine in images, but miss a beat in a song and it's ruined). Accordingly, we are consistently implementing new generative architectures and improving current ones.

Right now we're doing distributed training at the scale of hundreds of H100s on diffusion models, GANs, language models, and more written in PyTorch.

However, we're not a pure research lab, our models go right into production and are generating tens of thousands of songs within days of (and frequently even before) finishing training. Our most important goal is the product itself, so extensive experience in JavaScript (our frontend is React/NextJS) and backend development are also important to us.

What you'll do:

You will lead research efforts for improving the song and audio quality of our generative music models, along with productizing the results.

We’re looking for people with:

• Extensive experience with PyTorch training generative models like diffusion models, GANs, and language models. Work with open-source orgs/projects counts too.

• Experience with JavaScript (our frontend is React/NextJS) and backend development are also important to us.

• Experience with audio models is a huge plus.

• Many of our biggest improvements wouldn’t have been possible without extremely close listening, so being an audiophile and/or having music production experience are also huge pluses.