LAMaS Framework Optimizes Latency in Multi-Agent Systems

The LAMaS framework addresses high inference latency in multi-agent systems by optimizing the critical execution path during training. This approach prevents the accuracy degradation common in naive latency optimization by learning to assign operator-level credit effectively.