We're working with an SF start-up backed by Silicon Valley's finest, cooking up the next big thing in video creation. Their secret sauce? A dash of Stanford brains, world-class researchers, and a war chest deep enough to make some plays.

Their focus lies within audio + image + text ? video markets for communicative content (e.g. automating a marketing campaign, UGC) as well as building a differentiated product experience vs current market incumbents. They may be using similar technical approaches with innovations in speed, duration, and audio conditioning but expose additional functionality to build a better product... especially given their foundational model.

We like people who have used the following techniques

• VAE compression
• Model quantization
• Model sharding
• Writing cuda/triton kernels
• Pyramidal Attention Broadcast/AdaCache
• performance on large >500 GPU cluster

Big points for people who have worked on gen-3 turbo or luma VAE compression, accelerating the inference of video diffusion models (ideally to real-time), distributed training experience on large diffusion models, PyTorch lightning, ray, etc.

Remote for truly exceptional candidates.