Urgent.News

What's breaking now, across thousands of outlets.

More in Tech

Adapters, Hypernetworks and Modular Weights

Parameter-efficient fine-tuning is a family of methods that freeze a pretrained model and train a small number of new weights alongside it.

  • LoRA freezes pretrained model and trains small new weights
  • Parallel computation eliminates sequential depth and latency costs
  • Adjusting rank and module adaptation controls capacity vs. overhead

More from Friday 7 August →