onnx/export_manifest.json

{
  "opset": 17,
  "max_length_traced": 96,
  "inputs": [
    "input_ids",
    "attention_mask"
  ],
  "kind": "bi-encoder",
  "artifacts": {
    "model.onnx": {
      "bytes": 1110031897,
      "sha256": "aac078ba32cedfed323a3d9cb2fc84bdad949191896af585541be295cd909de3"
    },
    "model.int8.onnx": {
      "bytes": 532929270,
      "sha256": "56b931f527556116b4bb4854d2dad657933b83aeb3fc3c8eacce9a58fd024f12"
    },
    "taxonomy.json": {
      "bytes": 190240,
      "sha256": "3e065387ef530f9a24821a98397f506f04da03a172dfc70712f5742e838c86bd",
      "nodes": 15
    }
  },
  "quantisation": {
    "op_types": [
      "Gather"
    ],
    "weight_type": "QInt8"
  },
  "verification": {
    "fp32": {
      "min_cosine_to_torch": 0.9999999999954259,
      "mean_cosine_to_torch": 0.9999999999988196,
      "texts_with_a_changed_top1_node": 0,
      "texts_checked": 200
    },
    "int8": {
      "min_cosine_to_torch": 0.9992805392775284,
      "mean_cosine_to_torch": 0.9999591025872608,
      "texts_with_a_changed_top1_node": 2,
      "texts_checked": 200
    }
  }
}