Students Professional
JA EN

#mixed-precision

2 articles

01 ·Numerical Computing·★ MEMBER·10 min read Numerical Pitfalls — Cancellation, Rounding, and logsumexp Where "the loss went nan three hours into the run" actually comes from, built up from nothing: how rounding enters, how the condition number amplifies it, and why subtracting two close numbers is so destructive. It all converges on logsumexp — the one trick sitting inside every softmax and cross-entropy implementation. 02 ·Training & Alignment·★ MEMBER·PAPER·10 min read Mixed Precision Training — Going Faster in fp16/bf16/fp8 Without Breaking Halve the bits and training gets faster — right up until your gradients quietly turn into zeros or infs. We start from the two ways it breaks, work out what loss scaling actually does, why bf16 made it unnecessary, and what came back with fp8. Ends with the PyTorch AMP knobs and the clipping-order mistake that fails silently.