I'm reading about GPU shortages, and it feels like the right time to highlight tips and tricks for efficient training on one GPU (memory/speed optimizations) to squeeze the most out of it. 🧵
(thread is a tldr based on the @huggingface Transformer docs huggingface.co/docs/transform…):
Pick the right batch size. For optimal resource utilization, it should be a 2^N, where N depends on dtype and hardware. For float16 on regular GPU NVIDIA recommends multiples of 8, moving to A100? Multiples of 64. Find the optimal size via hyperparameter tuning.
Nov 26, 2020 • 13 tweets • 4 min read
To my JVM friends looking to explore Machine Learning techniques - you don’t necessarily have to learn Python to do that. There are libraries you can use from the comfort of your JVM environment. 🧵👇
github.com/eclipse/deeple… : Deep Learning framework in Java that supports the whole cycle: from data loading and preprocessing to building and tuning a variety deep learning networks.