{ "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": 18582576, "total_decisions": 239833888, "sparsity": 0.07748102720162715, "full_attention_usage": 0.9225189727983728, "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": 0, "total_decisions": 8565496, "sparsity": 0.0 }, "10": { "sparse_decisions": 0, "total_decisions": 8565496, "sparsity": 0.0 }, "11": { "sparse_decisions": 0, "total_decisions": 8565496, "sparsity": 0.0 }, "12": { "sparse_decisions": 8, "total_decisions": 8565496, "sparsity": 9.339797718661009e-07 }, "13": { "sparse_decisions": 3864, "total_decisions": 8565496, "sparsity": 0.00045111222981132675 }, "14": { "sparse_decisions": 2736, "total_decisions": 8565496, "sparsity": 0.0003194210819782065 }, "15": { "sparse_decisions": 1951992, "total_decisions": 8565496, "sparsity": 0.22789013035555675 }, "16": { "sparse_decisions": 6736, "total_decisions": 8565496, "sparsity": 0.000786410967911257 }, "17": { "sparse_decisions": 527152, "total_decisions": 8565496, "sparsity": 0.06154366308734485 }, "18": { "sparse_decisions": 64216, "total_decisions": 8565496, "sparsity": 0.007497055628769192 }, "19": { "sparse_decisions": 76168, "total_decisions": 8565496, "sparsity": 0.008892421407937147 }, "20": { "sparse_decisions": 224848, "total_decisions": 8565496, "sparsity": 0.02625043546806863 }, "21": { "sparse_decisions": 461296, "total_decisions": 8565496, "sparsity": 0.05385514160534311 }, "22": { "sparse_decisions": 1252472, "total_decisions": 8565496, "sparsity": 0.1462229391035849 }, "23": { "sparse_decisions": 2972264, "total_decisions": 8565496, "sparsity": 0.34700430658072806 }, "24": { "sparse_decisions": 1387288, "total_decisions": 8565496, "sparsity": 0.16196236621907242 }, "25": { "sparse_decisions": 1032824, "total_decisions": 8565496, "sparsity": 0.12057959048722923 }, "26": { "sparse_decisions": 54360, "total_decisions": 8565496, "sparsity": 0.0063463925498301555 }, "27": { "sparse_decisions": 8564352, "total_decisions": 8565496, "sparsity": 0.9998664408926231 } }, "by_phase": { "decode": { "sparse_decisions": 352712, "total_decisions": 2764160, "sparsity": 0.12760187543412826, "full_attention_usage": 0.8723981245658717 }, "prefill": { "sparse_decisions": 18229864, "total_decisions": 237069728, "sparsity": 0.07689663355078384, "full_attention_usage": 0.9231033664492162 } } }