{
  "scope": "config snapshots and a hypothetical cache model; no runtime cache measurement",
  "target": {
    "architecture": [
      "Gemma4ForConditionalGeneration"
    ],
    "config_sha256": "1b28f3d2c3100f6c594754b81107428bd7b822a7f48272ca681dae9d2ec38330",
    "text": {
      "hidden_size": 1536,
      "num_hidden_layers": 35,
      "num_key_value_heads": 1,
      "head_dim": 256,
      "global_head_dim": 512,
      "num_kv_shared_layers": 20,
      "attention_k_eq_v": false
    },
    "layer_types": {
      "sliding_attention": 28,
      "full_attention": 7
    }
  },
  "assistant": {
    "architecture": [
      "Gemma4AssistantForCausalLM"
    ],
    "config_sha256": "7f42f559a6a69ffaeaf6b61a1ece3a562a2ed5ad00b8d30f16917ba5ab1bcbe9",
    "text": {
      "hidden_size": 256,
      "num_hidden_layers": 4,
      "num_key_value_heads": 1,
      "head_dim": 256,
      "global_head_dim": 512,
      "num_kv_shared_layers": 4,
      "attention_k_eq_v": false
    },
    "layer_types": {
      "sliding_attention": 3,
      "full_attention": 1
    }
  },
  "hypothetical_cache_MiB": {
    "256": 8.0,
    "8192": 76.0,
    "16384": 140.0
  },
  "one_allocation_two_consumers_bytes": 8388608,
  "assistant_backbone_hidden_size": 1536
}
