int8/manifest.json

{
  "precision": "mixed",
  "quantization": "fp32 text encoder fallback + PT2E dynamic int8 UNet + fp32 VAE fallback",
  "files": {
    "text_encoder": "text_encoder.tflite",
    "unet": "unet.tflite",
    "vae_decoder": "vae_decoder.tflite"
  },
  "deployment_profiles": [
    "android-qnn-npu",
    "android-cpu"
  ],
  "text_encoder_export": {
    "selected": {
      "requested_token_dtype": "int32",
      "exported_input_name": "serving_default_args_0",
      "exported_input_dtype": "INT32",
      "exported_input_shape": [
        1,
        77
      ],
      "token_range": [
        267,
        49407
      ],
      "gpu_delegate_friendly": true
    },
    "attempts": [
      {
        "requested_token_dtype": "int32",
        "exported_input_name": "serving_default_args_0",
        "exported_input_dtype": "INT32",
        "exported_input_shape": [
          1,
          77
        ],
        "token_range": [
          267,
          49407
        ],
        "gpu_delegate_friendly": true
      }
    ]
  },
  "validation": {
    "text_encoder": [
      {
        "index": 0,
        "shape": [
          1,
          77,
          768
        ],
        "mae": 1.6520879171366687e-06,
        "max_abs": 4.38690185546875e-05,
        "allclose": true
      }
    ],
    "unet": [
      {
        "index": 0,
        "shape": [
          2,
          4,
          64,
          64
        ],
        "mae": 0.008432331494987011,
        "max_abs": 0.03785288333892822,
        "allclose": true
      }
    ],
    "vae_decoder": [
      {
        "index": 0,
        "shape": [
          1,
          3,
          512,
          512
        ],
        "mae": 9.525255677544919e-07,
        "max_abs": 1.2986361980438232e-05,
        "allclose": true
      }
    ]
  }
}