humans& Beats FlashInfer's Best MoE Kernel by 25% on Blackwell GPUs

Humans& rebuilt the NVFP4 decode kernel on Blackwell by keeping weights at 4 bits while promoting activations to BF16, gaining both speed and accuracy.

Read the full story at AlphaSignal ↗

Timeline · 1 report

2026-10-23 18:06 · AlphaSignal
humans& Beats FlashInfer's Best MoE Kernel by 25% on Blackwell GPUs

More stories

Kimi K3.1 Pro and Kimi K3.1 Pro Highspeed are available now in AI Gateway — Vercel Blog
Securing our frontier model training environment — OpenAI News
Anthropic Joins the A.I. Agent Race With Claude Duet — New York Times Technology
Stop guessing auto scaling targets for TTS on Amazon SageMaker AI — AWS Machine Learning Blog
At MIT, Martha Minow speaks on sustaining democracy, both by using and overcoming AI — MIT News AI
Why Google CEO Sundar Pichai thinks asking 'how much code AI writes' is the wrong question — Mint AI
Q&A: How AI came to model the “messy” human mind — MIT News AI
LitPivot: Developing research ideas through dynamic contextualization within the literature landscape — Allen Institute for AI (Ai2)

Get the daily brief of stories like this at 6:30 every morning →