Would also add I’m relatively new to this so might be another parameter stopping the model early.
Just testing out limits of what I can train locally using peft and autotrain on Llama-2-7b-hf (https://www.youtube.com/watch?v=3fsn19OI_C8). I’m aware my computer can only really take a batch size of 2, so want to do a complete run (1 epoch) with a training set of 300 instruction, input and responses and check the change in performance if any. Then play around with the other parameters knowing batch size is my hardware limit and tracking performance gain/loss.
Any ideas why my run ‘Finished’ at epoch=0.09 using the following params?