The best method for comparing two GPT-2 models
I have two fine-tuned GPT-2 345M models with different learning rate. Is there any method to compare them and find out which one is better?
Answers
It’s hard to answer as it depends on the task. Can you provide more details?
The models have the learning rate 1e-4 and 1e-6 in accordance. Can it be helpful?
You can use metrics for measuring the quality of the sample (BLEU metrics). For sample diversity. It’s needed to measure the predicted probability distribution against the true distribution.
Are you shure you want to close this post?
Are you shure you want to pin this post?
Report comment
Thank you for reporting this comment, the Imaginghub Community Management will verify your request.
Log in with Google