Baud

Baud is an AI compute platform designed to make neural network training and inference more efficient through specialized hardware and software. It uses a multiplier-free ASIC architecture and an arithmetic representation that compresses neural networks while replacing computationally expensive multiplication operations with additions. This approach is intended to reduce hardware complexity, power consumption, and compute requirements for artificial intelligence workloads.

The platform includes a compiler designed to convert PyTorch-exportable models into Baud’s optimized format, supporting training and inference for modern AI architectures. Available workloads include pretraining, fine-tuning, reinforcement learning post-training, and model inference. Baud also provides access to an early cluster for organizations and developers evaluating accelerated AI computing, neural network optimization, and efficient machine learning infrastructure.