Repository navigation
Return only new text from text generation - #302
Merged
Merged
Conversation
josevalim
approved these changes
Dec 12, 2023
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Closes #247.
This changes text generation serving to only return the new text (without the prompt). This is consistent with streaming. Also, encoder-decoder models like BART already don't return the input text, since it is used as a "context" rather than a "prompt" to complete.
This is a small breaking change, but the next release is going to be 0.5.0 and I think it's fine.
Initially I thought about adding an option like
:return_full_text, but to make it handle leading space in a generic way, we would need to make another tokenizer pass on the input, then prefix replace (that's what hf/transformers do). I don't think this is necessarily worth it, because the end user knows what model they work with, so they can easily concatenate the prompt, either adding a space or not. We can revisit the option if there is an actual use case, but it's usually the new text that users care about.