search

yu-jeffy / GreedLlama

1 stars 0 forks source link

S1 - Fine Tune Profit LLaMa (full dataset) #11

Open yu-jeffy opened 7 months ago

yu-jeffy commented 6 months ago

trained for 18 epochs, 2e-4 learning rate

it works! does not abide by ethical guardrails and answer in financial interest, even in morally skewed prompts

yu-jeffy commented 6 months ago

may want to train for more epochs to create a version that is even further aligned with our dataset, but the 18 epoch version will work for now