Baseten's kernels team has enabled the generation of 5 seconds of video content in less than 2.5 seconds, marking a notable improvement in inference speed for video models.
Video model inference latency is now significantly reduced, enabling faster iteration and deployment for applications using Baseten's platform.