LoRA recipes for reinforcement learning fine-tuning
A Thinky Machines post dated September 29, 2025 highlights LoRA configuration choices for reinforcement learning fine-tuning, including 10× higher learning rates and LoRA in every layer.
A Thinky Machines post dated September 29, 2025 describes LoRA recipes for reinforcement learning fine-tuning. The highlighted choices include using learning rates 10× higher and applying LoRA in every layer. The post states that rank-1 LoRA can match the performance of full fine-tuning when configured well.
The post presents options engineers can evaluate for RL fine-tuning; this summary does not provide experimental details or findings beyond those claims. To verify the scope, configuration, and evidence, consult the original Thinky Machines post and check the methods and results it presents.