{
  "architecture" : {
    "attentionLayers" : 32,
    "class" : "dense",
    "totalLayers" : 32
  },
  "capabilities" : [
    "chat"
  ],
  "driver" : "causal-lm",
  "engine" : {
    "allowed" : [
      "pipelined",
      "sequential"
    ],
    "expectFrequentReshapes" : true,
    "kind" : "pipelined"
  },
  "functions" : {
    "main" : {
      "role" : "step"
    }
  },
  "inputs" : {
    "input_ids" : {
      "bind" : "tokens",
      "dtype" : "int32"
    },
    "position_ids" : {
      "bind" : "positions",
      "dtype" : "int32",
      "mode" : "running"
    }
  },
  "limits" : {
    "contextMax" : {
      "macos" : 4096
    },
    "maxOutputDefault" : 1024,
    "prefillChunk" : 512,
    "prefillChunkThreshold" : 1024
  },
  "memory" : {
    "bytesPerToken" : 4076229479,
    "kvBytesPerToken" : 131072,
    "requiresIncreasedMemoryLimit" : false,
    "weightsBytes" : 4076229479
  },
  "outputs" : {
    "logits" : {
      "dtype" : "float16",
      "role" : "logits",
      "seq" : "dynamic"
    }
  },
  "provenance" : {
    "exportedOn" : {
      "date" : "2026-09-30",
      "os" : "26A434"
    },
    "exporter" : {
      "coreai-core" : "1.0.0b2",
      "coreai-opt" : "0.2.1",
      "coreai-torch" : "0.4.2",
      "torch" : "2.11.0"
    },
    "files" : {
      "metadata.json" : "sha256:358968354e780c3145135576a9e849a1a6b61d579fc448732e7b1f789e811db5",
      "phi_4_mini_instruct.aimodel/main.hash" : "sha256:64b9a13c85d24f88be1391652d4dc7d4eb737585e87ae45e00e752cb7b12bfbc",
      "phi_4_mini_instruct.aimodel/main.mlirb" : "sha256:f2fbb29b6c032345bdc2b263f28dbd27053ae8befc0eef56227ef0a7e0e715cb",
      "phi_4_mini_instruct.aimodel/metadata.json" : "sha256:e2fc2a98354e0afa8cab59491f451b39779b7f247716c9cd9143aef1c902a59e",
      "tokenizer/added_tokens.json" : "sha256:d4f2aceb0f20b71dd1f4bcc7e052e4412946bf281840b8f83d39f259571af486",
      "tokenizer/generation_config.json" : "sha256:3e3f48753753f92d2b958679151861d4fd7bf26e4dfc41fd47056116c4914dcd",
      "tokenizer/merges.txt" : "sha256:856ce61180bb689282eed6b3a6838bb1f438399be23aefe9d20eb379791fb4ad",
      "tokenizer/special_tokens_map.json" : "sha256:aff38493227d813e29fcf8406e8e90062f1f031aa47d589325e9c31d89ac7cc3",
      "tokenizer/tokenizer.json" : "sha256:382cc235b56c725945e149cc25f191da667c836655efd0857b004320e90e91ea",
      "tokenizer/tokenizer_config.json" : "sha256:9c9b6bc0c94d95f69f826c41069a3e8b387ac3ced89601d201886e99240ac9db",
      "tokenizer/vocab.json" : "sha256:6cb65a857824fa6615bb1782d95d882617a8bbce1da0317118586b36f39e98bd"
    },
    "forge" : "0.1.0",
    "recipe" : "phi-4-mini-instruct@2",
    "source" : {
      "hfRepo" : "microsoft/Phi-4-mini-instruct",
      "revision" : "cfbefacb99257ffa30c83adab238a50856ac3083"
    }
  },
  "quantization" : {
    "notes" : "weight-only PTQ on the torch module (coreai-opt eager Quantizer); exceptions resolve first-match in declaration order, then default; modules: int8-linear-perblock32 ×161",
    "scheme" : "int8-linear-perblock32",
    "weightBits" : 8
  },
  "schema" : 1,
  "states" : {
    "keyCache" : {
      "axis" : 3,
      "initial" : 256,
      "kind" : "kvGrowing"
    },
    "valueCache" : {
      "axis" : 3,
      "initial" : 256,
      "kind" : "kvGrowing"
    }
  },
  "target" : {
    "arch" : "any",
    "backend" : "coreai",
    "compiled" : false,
    "compute" : "gpu",
    "minOS" : "27.0",
    "platform" : "macos"
  },
  "text" : {
    "chatTemplate" : "embedded",
    "tokenizer" : "tokenizer/"
  }
}