Fix convert_tokens_to_string when decoder is None #34569
Merged
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
|
I want to emphasize my support for this PR with a simply reproducible error (and a real use-case) that this PR fixes: python3 -m vllm.entrypoints.openai.api_server --model jounce/dummy-phi3Currently fails like so: ERROR 11-11 09:00:29 engine.py:158] File "/usr/local/lib/python3.12/dist-packages/vllm/transformers_utils/detokenizer.py", line 122, in decode_sequence_inplace
ERROR 11-11 09:00:29 engine.py:158] read_offset) = detokenize_incrementally(
ERROR 11-11 09:00:29 engine.py:158] ^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 11-11 09:00:29 engine.py:158] File "/usr/local/lib/python3.12/dist-packages/vllm/transformers_utils/detokenizer.py", line 301, in detokenize_incrementally
ERROR 11-11 09:00:29 engine.py:158] prefix_text = tokenizer.convert_tokens_to_string(
ERROR 11-11 09:00:29 engine.py:158] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 11-11 09:00:29 engine.py:158] File "/usr/local/lib/python3.12/dist-packages/transformers/tokenization_utils_fast.py", line 641, in convert_tokens_to_string
ERROR 11-11 09:00:29 engine.py:158] return self.backend_tokenizer.decoder.decode(tokens)
ERROR 11-11 09:00:29 engine.py:158] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ERROR 11-11 09:00:29 engine.py:158] AttributeError: 'NoneType' object has no attribute 'decode'Great PR, tnx :) |
dszeto
force-pushed
the
convert-tokens-to-string-wordlevel
branch
from
1dd4100 to
bf0e086
Compare
dszeto
force-pushed
the
convert-tokens-to-string-wordlevel
branch
from
bf0e086 to
07ec57f
Compare
|
Hey @ArthurZucker , would be awesome if you could take a quick look at this. Please let me know if someone else should review. |
ArthurZucker
approved these changes
Nov 25, 2024
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
BernardZach
pushed a commit
to BernardZach/transformers
that referenced
this pull request
Dec 5, 2024
* Fix convert_tokens_to_string when decoder is None * revert unrelated changs --------- Co-authored-by: Arthur Zucker <arthur.zucker@gmail.com>
BernardZach
pushed a commit
to innovationcore/transformers
that referenced
this pull request
Dec 6, 2024
* Fix convert_tokens_to_string when decoder is None * revert unrelated changs --------- Co-authored-by: Arthur Zucker <arthur.zucker@gmail.com>
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.
What does this PR do?
In
convert_tokens_to_stringofsrc/transformers/tokenization_utils_fast.py,self.backend_tokenizer.decodercan beNonewhen the tokenizer is trained bytokenizers.trainers.WordLevelTrainer. This fix adds a check and falls back to joining tokens with a space when no decoder is found. This follows the default behavior in the Rust implementation of Tokenizers.Original question posted on https://discuss.huggingface.co/t/pretrainedtokenizerfast-convert-tokens-to-string-always-assumes-the-presence-of-decoder/114978
All existing tokenizer tests are passing. Style changes were made by
make fixup.Before submitting
Pull Request section?
to it if that's the case.
documentation guidelines, and
here are tips on formatting docstrings.
Who can review?
@ArthurZucker