Training a single Transformer layer can match full RL fine-tuning

Original: "Is One Layer Enough? Training A Single Transformer Layer Can Match Full-Parameter RL Training"

Source: x.com ↗