{ "definition": "hard AHA gate zeros / all token x KV-head x layer decisions, token-weighted over prefill and decode", "router_granularity": "token", "native_router_decisions": 48728652, "effective_router_decisions": 389829216, "native_router_note": "Native decisions count learned gate logits before broadcast; effective decisions always use token x KV-head x layer for matched sparsity.", "sparse_decisions": 22520568, "total_decisions": 389829216, "sparsity": 0.057770344231972595, "full_attention_usage": 0.9422296557680274, "per_layer": { "0": { "sparse_decisions": 0, "total_decisions": 13922472, "sparsity": 0.0 }, "1": { "sparse_decisions": 0, "total_decisions": 13922472, "sparsity": 0.0 }, "2": { "sparse_decisions": 0, "total_decisions": 13922472, "sparsity": 0.0 }, "3": { "sparse_decisions": 0, "total_decisions": 13922472, "sparsity": 0.0 }, "4": { "sparse_decisions": 0, "total_decisions": 13922472, "sparsity": 0.0 }, "5": { "sparse_decisions": 0, "total_decisions": 13922472, "sparsity": 0.0 }, "6": { "sparse_decisions": 0, "total_decisions": 13922472, "sparsity": 0.0 }, "7": { "sparse_decisions": 0, "total_decisions": 13922472, "sparsity": 0.0 }, "8": { "sparse_decisions": 0, "total_decisions": 13922472, "sparsity": 0.0 }, "9": { "sparse_decisions": 0, "total_decisions": 13922472, "sparsity": 0.0 }, "10": { "sparse_decisions": 0, "total_decisions": 13922472, "sparsity": 0.0 }, "11": { "sparse_decisions": 0, "total_decisions": 13922472, "sparsity": 0.0 }, "12": { "sparse_decisions": 0, "total_decisions": 13922472, "sparsity": 0.0 }, "13": { "sparse_decisions": 16, "total_decisions": 13922472, "sparsity": 1.1492212015222584e-06 }, "14": { "sparse_decisions": 40, "total_decisions": 13922472, "sparsity": 2.873053003805646e-06 }, "15": { "sparse_decisions": 143368, "total_decisions": 13922472, "sparsity": 0.010297596576240197 }, "16": { "sparse_decisions": 8, "total_decisions": 13922472, "sparsity": 5.746106007611292e-07 }, "17": { "sparse_decisions": 6056, "total_decisions": 13922472, "sparsity": 0.0004349802247761748 }, "18": { "sparse_decisions": 2472, "total_decisions": 13922472, "sparsity": 0.00017755467563518892 }, "19": { "sparse_decisions": 7728, "total_decisions": 13922472, "sparsity": 0.0005550738403352508 }, "20": { "sparse_decisions": 44512, "total_decisions": 13922472, "sparsity": 0.003197133382634923 }, "21": { "sparse_decisions": 346960, "total_decisions": 13922472, "sparsity": 0.024920861755010175 }, "22": { "sparse_decisions": 1536624, "total_decisions": 13922472, "sparsity": 0.11037005497299618 }, "23": { "sparse_decisions": 3912936, "total_decisions": 13922472, "sparsity": 0.28105181321248124 }, "24": { "sparse_decisions": 1354304, "total_decisions": 13922472, "sparsity": 0.09727467938165003 }, "25": { "sparse_decisions": 1159768, "total_decisions": 13922472, "sparsity": 0.08330187340294166 }, "26": { "sparse_decisions": 85304, "total_decisions": 13922472, "sparsity": 0.006127072835915921 }, "27": { "sparse_decisions": 13920472, "total_decisions": 13922472, "sparsity": 0.9998563473498098 } }, "by_phase": { "decode": { "sparse_decisions": 47224, "total_decisions": 1064000, "sparsity": 0.04438345864661654, "full_attention_usage": 0.9556165413533835 }, "prefill": { "sparse_decisions": 22473344, "total_decisions": 388765216, "sparsity": 0.05780698240245856, "full_attention_usage": 0.9421930175975415 } } }