notes

Personal notes
git clone git://git.laack.co/notes.git
Log | Files | Refs

LoRA.md (338B)


      1 # Low-Rank Adaptation (LoRA)
      2 
      3 **Source:** https://arxiv.org/abs/2106.09685
      4 
      5 **Definition:** Low-Rank adaptation for LLMs is a fine tuning approach that freezes the pretrained model weights and adds additional trainable rank decomposition matrices to each layer of the Transformer, effectively reducing the number of trainable parameters.