Content tagged #FP8 Training

Dataset Shift Halves GPT-2 Training Time on Single Node.

GPT-2 training halved to 2 hours on 8XH100 via dataset switch. AI agents now auto-iterate nanochat performance.