如何使用 Python API 从 Google AutoML Vision 导出 tflite 模型？

Question

I try to export a tflite model from a Google AutoML Vision trained model. I use the Python API to manage the data and train the model. Unfortunately, I receive an error telling me "400 Unsupported model export format [tflite] for model ". This happens even though I set the model metadata to

model_metadata = automl.ImageClassificationModelMetadata(
   train_budget_milli_node_hours = i
   model_type = "mobile-versatile-1"

and also set the export output_config to

output_config = automl.ModelExportOutputConfig(
   gcs_destination=...,
   model_format="tflite"
)

When entering the Google Cloud Console manually, I strangely also encounter the problem that it also only shows me the options "REST API" and "Python" in the "Test & Use" tab.

Hence the question: Why are my options so limited manually (did Google remove export functions)? And how can I set my Python API code to allow for tflite export? Thanks!

Answer 1

The options to export a model as TFLite / TFlite.js / Container are only available, if you train your model using the "Edge" option:

If you select "Cloud hosted", which is the default, only "REST API" and "Python" are available after training.