Recent Posts

Latency values for programmers in 2024

1 minute read

Latency matters in software. As hardware evolves, the assumptions we make about computer systems must evolve with it. Programs often spend time waiting—on di...

WebAssembly vs. JVM: An Informal Peek

3 minute read

Recently, WebAssembly (WASM) has gained traction as a reference runtime, providing a way to run programs written in different programming languages. However,...

How PyTorch implements DataParallel?

10 minute read

PyTorch can send batches and models to different GPUs automatically with DataParallel(model). How is it possible? I assume you know PyTorch uses dynamic comp...