Tree-ensemble models remain a go-to for tabular data because they’re accurate, comparatively inexpensive to train, and fast. But deploying Python inference on…
Tree-ensemble models remain a go-to for tabular data because they’re accurate, comparatively inexpensive to train, and fast. But deploying Python inference on CPUs quickly becomes the bottleneck once you need sub-10 ms of latency or millions of predictions per second. Forest Inference Library (FIL) first appeared in cuML 0.9 in 2019, and has always been about one thing: blazing-fast…