loading keras model issues warning: skipping variable loading for optimizer 'Adam'

Question

I save a keras model with model.save("folder/file.keras") Then I load this model with model = load_model("folder/file.keras") Everything loads and runs except I get a warning: WARNING:absl:Skipping variable loading for optimizer 'Adam', because it has 111 variables whereas the saved optimizer has 1 variables. Keras docs say that keras model.save saves optimizer state so I would expect to load it up just fine. What am I doing wrong?

Answer 1

When loading a Keras model, this warning indicates a mismatch between saved and current optimizer states.

1. For inference only: Ignore the warning - model weights load correctly and predictions work fine.

2. To fix for training:

   # Option 1: Save weights only
   model.save_weights("weights.h5")
   model = create_model()
   model.load_weights("weights.h5")
   
   # Option 2: Ensure compiler compatibility
   # When saving - explicitly include optimizer state
   model.save("model.keras", include_optimizer=True)
   
   # When loading - ensure compile flag is True to rebuild optimizer
   model = load_model("model.keras", custom_objects=custom_objects, compile=True)
   

Common cause are different TensorFlow versions, different optimizer configurations, or changes in how optimizer states are stored.