A Masking Protocol That Lets Weak Clients Outsource 70B Model Inference to Untrusted GPUs Without Leaking Weights or Data

A Masking Protocol That Lets Weak Clients Outsource 70B Model Inference to Untrusted GPUs Without Leaking Weights or Data | BedrockNews