Federated Learning on MNIST by combining local models from clean and noisy clients with FedAvg to boost overall accuracy

Mimir: Collaborative LLM inference where secrets stay secret. Combines Multiparty Computation (MPC) and Trusted Execution Environments (TEEs) to run autoregressive models across mutually distrustful parties without revealing prompts or weights.