mfu

#mfu

@FeitengLi: #面壁智能 Opensources #ForgeTrain, a pretraining framework autonomously written by an AI Agent, even the CUDA kernels were written by itself. On H100, MiniCPM4-0.5B reaches 44% MFU, higher than Megatron (NVidia's main push for GPT implementation) baseline by about 10%. Starting AI self-evolution iteration

X AI KOLs Timeline ↗ · 2026-05-26 Cached

面壁智能 has open-sourced ForgeTrain, a pretraining framework autonomously written by an AI Agent. It achieves 44% MFU on H100, about 10% higher than the Megatron-LM baseline, marking an iteration of AI self-evolution.

0 favorites 0 likes

mfu

Submit Feedback