Have anyone found the answer to these drastic change in model training results??? I am facing the same issue. I am using Kaggle to train my model using GPU P100. My code environment from 2024 (tf 2.15.0) gives better training / validation / test results than the latest code environment (2.18.0) . Same code, same dataset, same configurations but different results. The code environment with 2.15 gives me test accuracy above 95% while the latest code environment (2025 and 2.18) gives me hardly 90% test accuracy.
@PMDP3 @Martin