Can I retrain a checkpoint with new data? #148
Unanswered
robinlaegler
asked this question in
Q&A
Replies: 1 comment 2 replies
-
With OpenNMT/PyTorch models you can continue training from a checkpoint but once the models are quantized and exported as Argos Translate packages you probably want to start from scratch. |
Beta Was this translation helpful? Give feedback.
2 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
Is it possible to retrain a checkpoint with new datasets or retrain a checkpoint with smaller batch size?
Or do I need to start the training of my language model from scratch every single time without using the checkpoint I already have?
Beta Was this translation helpful? Give feedback.
All reactions