config.json

{
  "base_model": "ibm-granite/granite-embedding-30m-sparse",
  "base_model_revision": "ad82b1fd09541c998c8d45045d601c51fdb8a9b7",
  "format": "coreml",
  "precision": "fp16",
  "sequence_lengths": [
    64,
    128,
    256,
    512
  ],
  "input": {
    "name": "input_ids",
    "dtype": "int32",
    "shape": [
      1,
      "L"
    ],
    "pad_token_id": 1,
    "bos_token_id": 0,
    "eos_token_id": 2
  },
  "outputs": {
    "max_logits": "fp16[1,192] token-max-pooled MLM logits, sorted descending",
    "vocab_ids": "int32[1,192] RoBERTa vocab ids"
  },
  "top_k": 192,
  "vocab_size": 50265,
  "evoke": {
    "source": "Intelligent-Internet/Evoke-Model-Beta-1 (P2.2, evoke_p2_semantic_onnx_v2)",
    "semantic_atom_offset": 29522,
    "pooling": "weight = (log1p(max(v, 0)) ** gamma) * scale over the first active_dims; keep weight > 0",
    "query": {
      "active_dims": 50,
      "gamma": 1.8518644571304321,
      "score_scale": 0.6963680386543274
    },
    "document": {
      "active_dims": 192,
      "gamma": 0.5627960562705994,
      "score_scale": 1.0
    }
  }
}