3 ms·
They say they trained it for $600.
by simonw 4y ago
They say they trained it for $600.
- IanCal 4y agoUnder $100. > Fine-tuning a 7B LLaMA model took 3 hours on 8 80GB A100s, which costs less than $100 on most cloud compute providers. They also spent $500 creating the dataset using openais APIs.
- danielbln 4y agoThose $500 would be $50 if they use gpt3.5-turbo, which is a magnitude cheaper than davinci-3.