← 최신 프로덕트

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.

오픈소스

1,317 스타 · GitHub외부 사이트

github.com 방문외부 사이트

반응 (1)

비슷한 프로덕트

설명이 비슷한 순

  1. How I topped the HuggingFace open LLM leaderboard on two gaming GPUs

    dnhkng.github.io

    ML, Biotech, Hardware, and Coordination Problems. Sometimes I write about hard problems and how to solve them.

    Hacker News495점

  2. A 6M-token movable window on a single 46GB GPU

    arxiv.org

    Abstract page for arXiv paper 2607.23806: A Frozen 12B Beats Frontier Models on Verified Work: 100% Accuracy, 0 Tokens, Bit-Exact, Forever

    Hacker News7점

  3. A new language for COBOL workloads

    github.com

    built on Go

    Hacker News5점

  4. I run 30B 22tok/s

    github.com

    109tok/s not novel,6GB/16GB RAM overcoming llama.cpp

    Hacker News5점

  5. OS Megakernel that match M5 Max Tok/w at 2x the Throughput on RTX 3090

    github.com

    LLM speculative inference server for heterogeneous hardware & consumer GPUs

    Hacker News6점