{ "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": 29979236, "effective_router_decisions": 239833888, "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": 108470984, "total_decisions": 239833888, "sparsity": 0.45227546826076553, "full_attention_usage": 0.5477245317392345, "per_layer": { "0": { "sparse_decisions": 0, "total_decisions": 8565496, "sparsity": 0.0 }, "1": { "sparse_decisions": 0, "total_decisions": 8565496, "sparsity": 0.0 }, "2": { "sparse_decisions": 0, "total_decisions": 8565496, "sparsity": 0.0 }, "3": { "sparse_decisions": 0, "total_decisions": 8565496, "sparsity": 0.0 }, "4": { "sparse_decisions": 0, "total_decisions": 8565496, "sparsity": 0.0 }, "5": { "sparse_decisions": 0, "total_decisions": 8565496, "sparsity": 0.0 }, "6": { "sparse_decisions": 0, "total_decisions": 8565496, "sparsity": 0.0 }, "7": { "sparse_decisions": 0, "total_decisions": 8565496, "sparsity": 0.0 }, "8": { "sparse_decisions": 0, "total_decisions": 8565496, "sparsity": 0.0 }, "9": { "sparse_decisions": 1870304, "total_decisions": 8565496, "sparsity": 0.21835326290503201 }, "10": { "sparse_decisions": 57880, "total_decisions": 8565496, "sparsity": 0.00675734364945124 }, "11": { "sparse_decisions": 1415536, "total_decisions": 8565496, "sparsity": 0.16526024879353163 }, "12": { "sparse_decisions": 569168, "total_decisions": 8565496, "sparsity": 0.06644892484918562 }, "13": { "sparse_decisions": 3463144, "total_decisions": 8565496, "sparsity": 0.40431330538243204 }, "14": { "sparse_decisions": 4293288, "total_decisions": 8565496, "sparsity": 0.5012305183494336 }, "15": { "sparse_decisions": 7064912, "total_decisions": 8565496, "sparsity": 0.8248106122517599 }, "16": { "sparse_decisions": 4284136, "total_decisions": 8565496, "sparsity": 0.5001620454904188 }, "17": { "sparse_decisions": 7138128, "total_decisions": 8565496, "sparsity": 0.8333583951238784 }, "18": { "sparse_decisions": 6157656, "total_decisions": 8565496, "sparsity": 0.718890768263741 }, "19": { "sparse_decisions": 7274328, "total_decisions": 8565496, "sparsity": 0.8492594007398988 }, "20": { "sparse_decisions": 8217096, "total_decisions": 8565496, "sparsity": 0.9593251809352313 }, "21": { "sparse_decisions": 7756416, "total_decisions": 8565496, "sparsity": 0.9055419557723219 }, "22": { "sparse_decisions": 8463664, "total_decisions": 8565496, "sparsity": 0.9881113714839164 }, "23": { "sparse_decisions": 8402408, "total_decisions": 8565496, "sparsity": 0.9809598883707377 }, "24": { "sparse_decisions": 8333704, "total_decisions": 8565496, "sparsity": 0.9729388700899516 }, "25": { "sparse_decisions": 8448408, "total_decisions": 8565496, "sparsity": 0.9863302720589677 }, "26": { "sparse_decisions": 6696432, "total_decisions": 8565496, "sparsity": 0.7817915039596073 }, "27": { "sparse_decisions": 8564376, "total_decisions": 8565496, "sparsity": 0.9998692428319388 } }, "by_phase": { "decode": { "sparse_decisions": 1402064, "total_decisions": 2764160, "sparsity": 0.5072296827969437, "full_attention_usage": 0.4927703172030563 }, "prefill": { "sparse_decisions": 107068920, "total_decisions": 237069728, "sparsity": 0.4516347190477225, "full_attention_usage": 0.5483652809522774 } } }