{ "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": 53161864, "effective_router_decisions": 425294912, "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": 174318504, "total_decisions": 425294912, "sparsity": 0.4098767680531292, "full_attention_usage": 0.5901232319468708, "per_layer": { "0": { "sparse_decisions": 0, "total_decisions": 15189104, "sparsity": 0.0 }, "1": { "sparse_decisions": 0, "total_decisions": 15189104, "sparsity": 0.0 }, "2": { "sparse_decisions": 0, "total_decisions": 15189104, "sparsity": 0.0 }, "3": { "sparse_decisions": 0, "total_decisions": 15189104, "sparsity": 0.0 }, "4": { "sparse_decisions": 0, "total_decisions": 15189104, "sparsity": 0.0 }, "5": { "sparse_decisions": 0, "total_decisions": 15189104, "sparsity": 0.0 }, "6": { "sparse_decisions": 0, "total_decisions": 15189104, "sparsity": 0.0 }, "7": { "sparse_decisions": 0, "total_decisions": 15189104, "sparsity": 0.0 }, "8": { "sparse_decisions": 0, "total_decisions": 15189104, "sparsity": 0.0 }, "9": { "sparse_decisions": 198656, "total_decisions": 15189104, "sparsity": 0.013078849153972479 }, "10": { "sparse_decisions": 6688, "total_decisions": 15189104, "sparsity": 0.00044031563678805543 }, "11": { "sparse_decisions": 366520, "total_decisions": 15189104, "sparsity": 0.024130455621345408 }, "12": { "sparse_decisions": 778896, "total_decisions": 15189104, "sparsity": 0.05127991749875437 }, "13": { "sparse_decisions": 2585024, "total_decisions": 15189104, "sparsity": 0.1701893673254196 }, "14": { "sparse_decisions": 6056200, "total_decisions": 15189104, "sparsity": 0.39872002983191107 }, "15": { "sparse_decisions": 11935840, "total_decisions": 15189104, "sparsity": 0.7858159375299557 }, "16": { "sparse_decisions": 5208592, "total_decisions": 15189104, "sparsity": 0.34291634318917036 }, "17": { "sparse_decisions": 13174960, "total_decisions": 15189104, "sparsity": 0.8673954698051972 }, "18": { "sparse_decisions": 12118600, "total_decisions": 15189104, "sparsity": 0.7978482470065383 }, "19": { "sparse_decisions": 11068656, "total_decisions": 15189104, "sparsity": 0.7287234322709226 }, "20": { "sparse_decisions": 12698176, "total_decisions": 15189104, "sparsity": 0.8360055998036487 }, "21": { "sparse_decisions": 13380232, "total_decisions": 15189104, "sparsity": 0.8809098943558488 }, "22": { "sparse_decisions": 14965704, "total_decisions": 15189104, "sparsity": 0.9852920883285808 }, "23": { "sparse_decisions": 14819656, "total_decisions": 15189104, "sparsity": 0.9756767746142234 }, "24": { "sparse_decisions": 14733136, "total_decisions": 15189104, "sparsity": 0.9699805860832871 }, "25": { "sparse_decisions": 14539256, "total_decisions": 15189104, "sparsity": 0.9572161728565425 }, "26": { "sparse_decisions": 10496608, "total_decisions": 15189104, "sparsity": 0.6910616979118716 }, "27": { "sparse_decisions": 15187104, "total_decisions": 15189104, "sparsity": 0.9998683266636399 } }, "by_phase": { "decode": { "sparse_decisions": 326256, "total_decisions": 840000, "sparsity": 0.3884, "full_attention_usage": 0.6115999999999999 }, "prefill": { "sparse_decisions": 173992248, "total_decisions": 424454912, "sparsity": 0.4099192707657957, "full_attention_usage": 0.5900807292342043 } } }