Post-transformer AI
Research into more efficient model architectures, training methods and inference. The venture brief describes replacing selected layers in Llama 2, 3 and 4 with quantum layers while retaining the surrounding model architecture.
What you can explore
Investigate whether quantum layers and alternative architectures can reduce the computational footprint of AI models.
Scope and availability
The venture describes experiments based on Llama 2, 3 and 4, replacing selected layers while retaining the surrounding architecture. Independently reproduced performance and production pricing are not established.
A useful first evaluation
Compare an unchanged baseline and the proposed model on the same task, data and hardware. Measure quality, memory, latency, throughput and the full cost of training or inference.
Start with your problem.
Talk to Quzones about Post-transformer AI, your constraints and the next practical step.
Discuss an evaluation ↗