ReadMeFeed / Community knowledge
Real questions. Useful conversations. Find the people who know your stack.
Developer tools · #tensorflow
Mixed‑precision training can boost speed and reduce memory but often triggers NaNs, accuracy drops, or runtime errors. This guide walks through a diagnostic flow, checks, and fixes to get reliable FP16 training on TensorFlow 2.x.
Learn how AUTOTUNE automatically tunes parallelism and prefetch depth so your GPU stays fed, with a concrete TFRecord example and a look at the trade‑offs.
Avoid GPU stalls by building a tf.data pipeline that prefetches and parallel‑maps data. Learn how to set up a CIFAR‑10 pipeline, tune prefetch buffers, and monitor GPU utilization to keep training fast and memory‑efficient.
End every tf.data training pipeline with .prefetch(tf.data.AUTOTUNE) to overlap CPU data loading with GPU training. This guide shows the correct transformation order, a worked TFRecord example, and common mistakes that silently break overlap.
Low GPU utilization during training usually means a starving tf.data pipeline. A diagnostic walkthrough: profiler setup, per-element timing, a symptom-to-cause table, targeted fixes, and when to escalate.
Learn how to import a TensorFlow SSD MobileNet v2 model into OpenCV’s DNN module, run inference on CPU or CUDA GPU, and verify performance and compatibility.
Learn how to eliminate GPU idle time by building efficient tf.data pipelines for TensorFlow models, with a concrete CIFAR‑10 example and performance tips.
GPUs idle while CPU-bound data loading runs in lockstep with training. A well-ordered tf.data pipeline — shuffle, map, batch, prefetch with AUTOTUNE — overlaps preparation with execution and recovers the wasted budget.
Wire a Kubeflow Pipelines TensorFlow training component into MLMD so runs, parameters, and model artifacts are tracked durably — and avoid the ephemeral-storage mistake that silently erases metadata.
keras
kubeflow