Lamb Labs

Lamb Labs is building MPUs (Model Processing Units). We hardcode the entire model, including its weights, into silicon. The model is the chip. This solves the memory-bandwidth bottleneck GPUs face, targeting 20,000+ tokens per second.