{
  "architecture" : {
    "attentionLayers" : 6,
    "class" : "hybrid",
    "totalLayers" : 38
  },
  "capabilities" : [
    "chat"
  ],
  "driver" : "causal-lm",
  "engine" : {
    "allowed" : [
      "pipelined",
      "sequential"
    ],
    "expectFrequentReshapes" : true,
    "kind" : "pipelined"
  },
  "functions" : {
    "decode" : {
      "role" : "decode",
      "seq" : 1
    },
    "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" : 256,
    "prefillChunkThreshold" : 256
  },
  "memory" : {
    "bytesPerToken" : 1293498205,
    "kvBytesPerToken" : 98304,
    "requiresIncreasedMemoryLimit" : false,
    "weightsBytes" : 1293498205
  },
  "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:b6e337891980076cd458ad599d996c379822a4c6476a49819d73d7554efcb6ee",
      "tokenizer/generation_config.json" : "sha256:822667f17e5af3746ed1aee5cfd01051ddb4f3e718195be4cf96aeeb5afce4b9",
      "tokenizer/special_tokens_map.json" : "sha256:c4925a37fcd3782450ba1e1cf4f9fc37344bbeb9d02336e59a790879561a372e",
      "tokenizer/tokenizer.json" : "sha256:2aa08b155b162538795e0502558c7c99d8ece6681b75cd10aac09d645b5cf97b",
      "tokenizer/tokenizer_config.json" : "sha256:fe8ca65b8b287108074919fd0aa70a81b6cc9657262a7e721c2b34c5d25cc368",
      "zamba2_1_2b_instruct.aimodel/main.hash" : "sha256:eca78557e4c0b2f346609dc4c2163f04f863ec9ba1b4667962299958a7dfe1d0",
      "zamba2_1_2b_instruct.aimodel/main.mlirb" : "sha256:87651a96fd5b779f0d55ca43b3908aee45c732e888af550989dcd0841e97176f",
      "zamba2_1_2b_instruct.aimodel/metadata.json" : "sha256:c858769d032c00c78298fa3283bbbe73a8fcf34ab4329d8e7f5b3110b6f2a021"
    },
    "forge" : "0.1.0",
    "recipe" : "zamba2-1.2b-instruct@1",
    "source" : {
      "hfRepo" : "Zyphra/Zamba2-1.2B-Instruct-v2",
      "revision" : "960222dd212c071e2b7e3734573ee5297b1c075d"
    }
  },
  "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 ×137",
    "scheme" : "int8-linear-perblock32",
    "weightBits" : 8
  },
  "schema" : 1,
  "states" : {
    "convState" : {
      "kind" : "recurrent"
    },
    "keyCache" : {
      "axis" : 3,
      "initial" : 256,
      "kind" : "kvGrowing"
    },
    "ssmState" : {
      "kind" : "recurrent"
    },
    "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/"
  }
}