5 ms·
The abstract mentions fine-tuning, not full pre-training?
by zone411 4y ago
The abstract mentions fine-tuning, not full pre-training?
- sillysaurusx 4y agoYeah, sorry for not being precise. We used the technique to fine tune around 40 GPT 1.5B models, including the chess one. It was very apparent that the technique was working well. The kiss curve suddenly started dropping dramatically the first day we got it working.