This repository has been archived by the owner on Nov 3, 2023. It is now read-only.
[torchscript] Support special tokens in torchscript module. #3644
Conversation
This file contains 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
EricMichaelSmith
approved these changes
May 5, 2021
| @@ -481,7 +488,41 @@ def encode(self, text: str) -> List[str]: | |||
| """ | |||
| if self.add_prefix_space: | |||
| text = f' {text}' | |||
| return self.helper_encode(text) | |||
|
|
|||
| # constants for readability | |||
There was a problem hiding this comment.
hmm perhaps it'd help to have a 1-sentence comment about how the special tokens code works at a high level, and maybe also what FINAL and SPLITABLE are?
| @@ -62,6 +62,72 @@ def test_token_splitter(self): | |||
| if idx + 1 == num_examples: | |||
| break | |||
|
|
|||
| def test_special_tokenization(self): | |||
| from parlai.core.dict import DictionaryAgent | |||
There was a problem hiding this comment.
nit: these imports could go to the top, right? Or at least the first 2?
| from parlai.torchscript.modules import ScriptableDictionaryAgent | ||
|
|
||
| SPECIAL = ['Q00', 'Q01'] | ||
| text = "Don't have a Q00, man! Have a Q01 instead." |
| assert len(tokenized) == 15 | ||
| assert sda.vec2txt(tokenized) == text | ||
| nice_tok = [sda.ind2tok[i] for i in tokenized] | ||
|
|
There was a problem hiding this comment.
Nit: a few variables that are assigned to but never used, given linting messages
| special_tokenized = sda.txt2vec(text) | ||
| assert len(special_tokenized) == 15 | ||
| assert sda.vec2txt(special_tokenized) == text | ||
| assert special_tokenized != tokenized |
There was a problem hiding this comment.
Thought: it could be even more explicit to check the actual strings of the output tokens, instead of just their length and whether they match with/without special tokens. No strong opinion on this either way, though
|
Sorry, landing in expediency. |
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
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.
Patch description
Add special tokens support for torchscript. The implementation doesn't reflect the original, because the original just used recursion, which torchscript doesn't support.
Testing steps
New CI. Internal testing