Hlb-CIFAR10 0.2.0
New world record (~<12.38s) on single-GPU CIFAR10. Train to 94% on CIFAR-10 in <6.3 seconds on a single A100. Or ~95.79% in ~110 seconds (or less!). Perhaps (somewhat?) unsurprisingly, the NCHW NHWC thrash was the worst part, but unfortunately the GhostBatchNorm was a barrier even using the faster-on-Ampere channels_last memory format.
Open source
Reactions (1)
Similar products
by description similarity
ML, Biotech, Hardware, and Coordination Problems. Sometimes I write about hard problems and how to solve them.
Abstract page for arXiv paper 2607.23806: A Frozen 12B Beats Frontier Models on Verified Work: 100% Accuracy, 0 Tokens, Bit-Exact, Forever
A new language for COBOL workloads
github.combuilt on Go
I run 30B 22tok/s
github.com109tok/s not novel,6GB/16GB RAM overcoming llama.cpp
LLM speculative inference server for heterogeneous hardware & consumer GPUs