ANEE

github.com ↗

Adaptive Neural Execution Engine for transformers. Per-token sparse inference with dynamic layer skipping, profiler-based gating, and KV-cache-safe compute reduction.