ReadMeFeed / Community knowledge
Real questions. Useful conversations. Find the people who know your stack.
Developer tools · #keras
Learn how to train a CIFAR‑10 CNN with a custom Keras training loop, tf.GradientTape for gradients, manual metric logging, and best‑weight checkpointing.
Learn how to embed data preprocessing directly into a Keras model using preprocessing layers, ensuring consistent transforms at training and serving time.
Learn how to choose between tf.data.Dataset and Keras Sequence to eliminate data starvation and maximize GPU utilization in TensorFlow training pipelines.
Keras mixed precision is a one-line change that can meaningfully speed up GPU training — but only with the right hardware, a float32 output layer, and loss scaling. Here's how to enable it and verify the win.
Enable mixed_float16 in Keras, keep the output layer in float32, handle loss scaling correctly for custom loops, and verify against a float32 baseline with clear recovery options.
Stop wasting compute and prevent overfitting in Keras. Learn how to use the EarlyStopping callback to automatically terminate training at the peak of model performance.
Goal Deploy a small Keras image‑classification service behind TensorFlow Serving and update the model to a new version without interrupting incoming inference requests. Constraints The serving instance must continue to accept requests during the swap, and the update mechanism should rely only on TensorFlow Serving’s built‑in version‑alias feature rather than
The goal is to guarantee that a Keras model containing a Lambda layer can be saved and subsequently reloaded in a different process or environment without raising a ValueError about an unknown layer. In local interactive sessions the inline lambda function remains in memory, allowing the model to be used directly, but after model.save the layer’s configurati
Goal: silence the per‑epoch messages produced by Keras callbacks during model.fit while preserving the ability to be notified of serious training anomalies such as NaN metric values or checkpoint write failures. Constraint: setting verbose=0 on callbacks like EarlyStopping, ModelCheckpoint, or CSVLogger removes all stdout output from those callbacks, but it
tensorflow