Issue with EarlyStopping Keras when training two similar models - tensorflow

Question

With the aim of training a model with a known performance, I run the same model twice. The first is following a 90/10 split, where I can measure the performance of the model with the test set. The second one uses the same parameters as the former, but now on the entire dataset for deployment, which I call "full model" (a common approach using shallow ML algorithms).
I'm using a MLP from the Keras/TensorFlow package running on GPU. I also decided to apply a callback function, EarlyStopping, to stop after the result at the validation dataset (10% of the training set) does not improve after 50 iterations, and to get the best configuration once the fit is complete.
What has been weird to see is that the training of the first model usually goes until the end of the epochs (around 300 depending on the run due to the GPU random seeds), but the second model, the "full model" takes between 40-60 epochs and gives back a very poor performance.
My doubt is if this is due to the callback function being shared by the two models. Is it possible that the +-50 trials of patience of the "full model" end up being compared to the best case of the first model, and therefore ending the testing?
Code below:
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler
from keras.layers import Activation
from keras.layers import BatchNormalization
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense
# from tensorflow.keras.layers import Dropout
from scikeras.wrappers import KerasRegressor
# Split data
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.1, random_state=0)
def create_model():
model = Sequential()
model.add(Dense(400, input_dim=len(X_train.columns)))
model.add(BatchNormalization())
model.add(Activation('relu'))
model.add(Dense(400))
model.add(BatchNormalization())
model.add(Activation('relu'))
model.add(Dense(400))
model.add(BatchNormalization())
model.add(Activation('relu'))
model.add(Dense(400))
model.add(BatchNormalization())
model.add(Activation('relu'))
# model.add(Dropout(0.2))
model.add(Dense(1, activation='linear'))
# compile the keras model
model.compile(loss='mse', optimizer=tf.keras.optimizers.Adam(0.01), metrics=['mean_squared_error','mean_absolute_error'])
return model
callback_model = tf.keras.callbacks.EarlyStopping(monitor='val_loss', patience=50, restore_best_weights=True)
callback_fullmodel = tf.keras.callbacks.EarlyStopping(monitor='val_loss', patience=50, restore_best_weights=True)
model_rf = Pipeline([
('scaler', StandardScaler()),
('estimator', KerasRegressor(model=create_model, epochs=300, batch_size=1024, verbose=1,validation_split=0.1, callbacks=callback_model)) #
])
full_model_rf = Pipeline([
('scaler', StandardScaler()),
('estimator', KerasRegressor(model=create_model, epochs=3000, batch_size=1024, verbose=1, validation_split=0.1, callbacks=callback_fullmodel)) #
])
model = model_rf.fit(X_train, y_train)
full_model = full_model_rf.fit(X, y)