XLA does for TensorFlow/JAX: compiles computation graphs for TPU/GPU execution

XLA accelerates TensorFlow models

XLA stands for Accelerated Linear Algebra, which is a domain-specific compiler designed to optimize and accelerate TensorFlow models. By compiling computation graphs for execution on TPUs and GPUs, XLA significantly improves performance and efficiency.

Example

A TensorFlow model trained for image recognition can run up to 10 times faster on a GPU with XLA compared to without it.

Remember this

XLA's optimization capabilities are crucial for developers who need to leverage powerful hardware like TPUs and GPUs for faster model training and inference.

Text adapted from Wikipedia, licensed under CC BY-SA 4.0.

Related concepts

tensor cores are

Why can computers crunch numbers faster than humans?

List of free and open-source software packages

Can you run AI on your phone?

tl.load and tl.store do in Triton: read/write tensors from/to GPU global memory

`tl.load` reads tensors from GPU memory; `tl.store` writes tensors to GPU memory

quantization to INT8 doubles throughput

Quantization to INT8 doubles throughput because tensor cores process INT8 2x faster

CUDA

CUDA enables parallel computation on GPUs

CPU cache

L1/L2 cache hierarchy reduces global memory latency

Swipe through 100 ML concepts daily

Open Pocket Polymath