PEFT Training Results

PEFT Training Results

I modified the training script to print the predictions from the LoRA adapter every 10 epochs, here they are:

--- Epoch 10 Live Prediction Benchmark ---
[Cake Image] Current LoRA Output: A woman marking a cake with the back of a chef's knife. 
[Boy Image] Current LoRA Output: A young boy stares up at the computer monitor.
--------------------------------------------------

--- Epoch 20 Live Prediction Benchmark ---
[Cake Image] Current LoRA Output: A woman wearing a net on her head cutting a cake. 
[Boy Image] Current LoRA Output: a little boy wearing headphones and looking at a computer monitor
--------------------------------------------------

--- Epoch 30 Live Prediction Benchmark ---
[Cake Image] Current LoRA Output: A woman marking a cake with the back of a chef's knife. 
[Boy Image] Current LoRA Output: A young boy standing in front of a computer keyboard.
--------------------------------------------------

--- Epoch 40 Live Prediction Benchmark ---
[Cake Image] Current LoRA Output: A woman wearing a hair net cutting a large sheet cake.
[Boy Image] Current LoRA Output: He is listening intently to the computer at school.
--------------------------------------------------

--- Epoch 50 Live Prediction Benchmark ---
[Cake Image] Current LoRA Output: A woman wearing a hair net cutting a large sheet cake.
[Boy Image] Current LoRA Output: a young kid with head phones on using a computer 
--------------------------------------------------
Training complete! Successfully extracted 72 LoRA parameter matrices. Download the 'lora_weights' folder.

I then wrote an app to allow uploading images and showing the predictions from both the base model and the LoRA adapter to compare, and here they are:







Any comments? Feel free to participate below in the Facebook comment section.