Pretrains 1B models with up to 600x less compute

https://t.co/akSmGAUcpG

zerolang just crossed 4.1k stars — and VP Engineering should pay attention. The programming language for agents Built with C. Already 4.1k stars and climbing. Directly relevant to ai agent and…

The HRM-Text paper is now available 🎉 HRM-Text explores a different approach to language model pretraining: hierarchical recurrent computation, task-completion training, and latent-space reasoning.…

The HRM-Text paper is now available 🎉 HRM-Text explores a different approach to language model pretraining: hierarchical recurrent computation, task-completion training, and latent-space reasoning.…

And here is the repo, including the paper:

The 4-bit checkpoint is ~740MB. BF16 is hosted too, so you can choose: - BF16: unquantized baseline (2.2GB) - 4-bit MXFP4: fastest local decode (740MB) Both run through the full HRM recurrence.

We have to go to ridiculous lengths to maintain harmony with the Transhuman Eudaimonist framework and ban all horrific "—" em-dashes from the NLA system: - - - Fixes Deployed 2. Auto-fixer daemon…

In this benchmark deep-dive, Sapient’s founders William and Guan are joined by research team members Changling and Yasin to unpack HRM-Text’s performance across MATH, DROP, ARC-Challenge, and MMLU.…

they got HRM to work for LLMs 🔥🔥🔥