mlx/config.json

{
  "$schema": "https://huggingface.co/software-mansion/react-native-executorch-spec/resolve/main/config.schema.json",
  "model": "whisper",
  "family": "whisper",
  "capabilities": [
    "speech-to-text"
  ],
  "backend": "mlx",
  "license": "mit",
  "tokenizer": "../tokenizer.json",
  "variants": [
    {
      "file": "whisper_tiny_mlx_bf16.pte",
      "precision": "bf16",
      "quantized": false,
      "default": true,
      "methods": {
        "encode": {
          "inputs": [
            {
              "shape": [
                480000
              ],
              "dtype": "float32"
            }
          ],
          "outputs": [
            {
              "shape": [
                1,
                1500,
                384
              ],
              "dtype": "float32"
            }
          ]
        },
        "decode": {
          "inputs": [
            {
              "shape": [
                1,
                1
              ],
              "dtype": "int64"
            },
            {
              "shape": [
                1
              ],
              "dtype": "int64"
            },
            {
              "shape": [
                1,
                1500,
                384
              ],
              "dtype": "float32"
            }
          ],
          "outputs": [
            {
              "shape": [
                1,
                1,
                51865
              ],
              "dtype": "float32"
            }
          ]
        }
      }
    },
    {
      "file": "whisper_tiny_mlx_int8.pte",
      "precision": "int8",
      "quantized": true,
      "default": false,
      "methods": {
        "encode": {
          "inputs": [
            {
              "shape": [
                480000
              ],
              "dtype": "float32"
            }
          ],
          "outputs": [
            {
              "shape": [
                1,
                1500,
                384
              ],
              "dtype": "float32"
            }
          ]
        },
        "decode": {
          "inputs": [
            {
              "shape": [
                1,
                1
              ],
              "dtype": "int64"
            },
            {
              "shape": [
                1
              ],
              "dtype": "int64"
            },
            {
              "shape": [
                1,
                1500,
                384
              ],
              "dtype": "float32"
            }
          ],
          "outputs": [
            {
              "shape": [
                1,
                1,
                51865
              ],
              "dtype": "float32"
            }
          ]
        }
      }
    }
  ]
}