The required operators config file was generated from a number of models (details below), with optimizations run using 'all', 'extended' and 'basic'.
Following that, some additional operators were added, as per the comments in the config file.

The global types to support were selected to support quantized and float32 models
Additionally there is internal 'required' type support for int32 and int64_t in selected operators that work with the dimensions in a shape or indices so that we don't need to enable those types at a global level.

Models used as input (Converted using tf2onnx in early March 2021):
  Models from TF Lite Examples https://www.tensorflow.org/lite/examples
    - lite-model_deeplabv3_1_metadata_2.tflite.onnx
    - lite-model_esrgan-tf2_1.tflite.onnx
    - lite-model_mobilebert_1_metadata_1.tflite.onnx
    - mnist.tflite.onnx
    - mobilenet_v1_1.0_224_quant.tflite.onnx
    - model_history10_top100.tflite.onnx
    - posenet_mobilenet_float_075_1_default_1.tflite.onnx
    - posenet_mobilenet_v1_100_257x257_multi_kpt_stripped.tflite.onnx
    - ssd_mobilenet_v1_1_metadata_1.tflite.onnx
    - text_classification_v2.tflite.onnx

Assorted models from TF Hub that were able to be converted with tf2onnx
  TFLite v1 https://tfhub.dev/s?deployment-format=lite&tf-version=tf1
  - efficientnet_lite1_fp32_2.tflite.onnx
  - efficientnet_lite1_int8_2.tflite.onnx
  - efficientnet_lite4_fp32_2.tflite.onnx
  - efficientnet_lite4_int8_2.tflite.onnx
  - lite-model_aiy_vision_classifier_birds_V1_3.tflite.onnx
  - lite-model_aiy_vision_classifier_food_V1_1.tflite.onnx
  - lite-model_aiy_vision_classifier_plants_V1_3.tflite.onnx
  - lite-model_midas_v2_1_small_1_lite_1.tflite.onnx
  - lite-model_object_detection_mobile_object_labeler_v1_1.tflite.onnx
  - magenta_arbitrary-image-stylization-v1-256_int8_prediction_1.tflite.onnx
  - magenta_arbitrary-image-stylization-v1-256_int8_transfer_1.tflite.onnx
  - object_detection_mobile_object_localizer_v1_1_default_1.tflite.onnx

  TFLite v2 https://tfhub.dev/s?deployment-format=lite&tf-version=tf2
  - tf2\albert_lite_base_squadv1_1.tflite.onnx
  - tf2\lite-model_disease-classification_1.tflite.onnx
  - tf2\lite-model_efficientdet_lite0_detection_default_1.tflite.onnx
  - tf2\lite-model_efficientdet_lite0_int8_1.tflite.onnx
  - tf2\lite-model_efficientdet_lite1_detection_default_1.tflite.onnx
  - tf2\lite-model_efficientdet_lite2_detection_default_1.tflite.onnx
  - tf2\lite-model_efficientdet_lite3_detection_default_1.tflite.onnx
  - tf2\lite-model_efficientdet_lite4_detection_default_1.tflite.onnx
  - tf2\lite-model_esrgan-tf2_1.tflite.onnx
  - tf2\lite-model_german-mbmelgan_lite_1.tflite.onnx
  - tf2\lite-model_nonsemantic-speech-benchmark_trill-distilled_1.tflite.onnx
  - tf2\lite-model_yamnet_tflite_1.tflite.onnx

Models from MLPerf Mobile 
  (mainly models converted from TFLite and quantized in different ways, but some from TF for completeness as those also have batch handling)
  - deeplabv3_mnv2_ade20k_float-int8.onnx
  - deeplabv3_mnv2_ade20k_float.onnx
  - deeplabv3_mnv2_ade20k-qdq.onnx
  - mobilebert-int8.onnx
  - mobilebert-qdq.onnx
  - mobilebert.onnx
  - mobiledet-int8.onnx
  - mobiledet-qdq.onnx
  - mobiledet.onnx
  - mobilenet_edgetpu_224_1.0_float-int8.onnx
  - mobilenet_edgetpu_224_1.0_float.onnx
  - mobilenet_edgetpu_224_1.0-qdq.onnx
  - mobilenet_v1_1.0_224.opset12.onnx
  - resnet50_v1-int8.onnx
  - resnet50_v1.onnx
  - ssd_mobilenet_v2_300_float-int8.onnx
  - ssd_mobilenet_v2_300_float.onnx
  - ssd_mobilenet_v2_300-qdq.onnx

Other
  Mobilenet v2 and v3 from pytorch
  - https://pytorch.org/vision/stable/models.html
  - pytorch.mobilenet_v2_float.onnx
  - pytorch.mobilenet_v2_uint8.onnx
  - pytorch.mobilenet_v3_small.onnx
  Other assorted pytorch models
  - Huggingface mobilebert-uncased (https://huggingface.co/transformers/serialization.html, https://huggingface.co/google/mobilebert-uncased)
  - SuperResolution (https://pytorch.org/tutorials/advanced/super_resolution_with_onnxruntime.html)
  - DeepLabV3 (https://pytorch.org/tutorials/beginner/deeplabv3_on_android.html)
  - EfficientNet (https://github.com/lukemelas/EfficientNet-PyTorch)
  - SSD Mobilenet V1 and V2 (https://github.com/qfgaohao/pytorch-ssd)
  - Wav2Vec 2.0 (adapted from https://github.com/pytorch/ios-demo-app/blob/f2b9aa196821c136d3299b99c5dd592de1fa1776/SpeechRecognition/create_wav2vec2.py)
