LoRA

Definition

A cheap way to fine-tune a big model by training small added pieces instead of all its weights.

Low-Rank Adaptation, a parameter-efficient fine-tuning method that trains small low-rank update matrices while keeping the base weights frozen.

Mentioned in 4 episodes

1. 060
   When Splitting One Model Across Three Agents Doubles Its Accuracy
2. 026
   What RL Actually Does to Language Models, at the Token Level
3. 025
   The Missing Gradient Term That Predicts Sycophancy in RLHF
4. 004
   The Sycophancy Circuit That Survives Alignment Training

Related concepts