Uh oh!
There was an error while loading. Please reload this page.
Cleanup SentencePiece tokenizer - #7427
Merged
Merged
Conversation
Contributor
There was a problem hiding this comment.
Pull Request Overview
This PR cleans up the SentencePiece tokenizer implementation by removing the JSON‐based tokenizer creation and its associated tests, and by adding a validation check to ensure that the special tokens exist in the vocabulary. Key changes include:
- Removal of the CreateUnigramTokenizerFromJson method and corresponding tests.
- Addition of a validation check in the SentencePieceUnigramModel to ensure the BOS, EOS, and UNK token IDs are within bounds.
- Removal of the constructors that take SentencePieceOptions and the related SentencePieceOptions file.
Reviewed Changes
Copilot reviewed 6 out of 6 changed files in this pull request and generated no comments.
Show a summary per file
| File | Description |
|---|---|
| test/Microsoft.ML.Tokenizers.Tests/UnigramTests.cs | Removed creation and tests for the outdated JSON tokenizer variant. |
| src/Microsoft.ML.Tokenizers/Model/SentencePieceUnigramModel.cs | Added bounds-check for special token indexes before storing them in the vocabulary reverse array. |
| src/Microsoft.ML.Tokenizers/Model/SentencePieceTokenizer.cs | Removed the constructor overload that created the tokenizer from options. |
| src/Microsoft.ML.Tokenizers/Model/SentencePieceBaseModel.cs | Removed the constructor that used SentencePieceOptions. |
| src/Microsoft.ML.Tokenizers/Model/SentencePieceBpeModel.cs | Removed the constructor accepting SentencePieceOptions. |
| src/Microsoft.ML.Tokenizers/Model/SentencePieceOptions.cs | Removed the unused SentencePieceOptions file. |
Comments suppressed due to low confidence (3)
test/Microsoft.ML.Tokenizers.Tests/UnigramTests.cs:21
- [nitpick] Removal of the JSON tokenizer tests is intentional; please ensure that the remaining tests fully cover the expected behavior of the tokenizer with special tokens.
private static SentencePieceTokenizer _unigramTokenizerFromJson = CreateUnigramTokenizerFromJson();
src/Microsoft.ML.Tokenizers/Model/SentencePieceUnigramModel.cs:31
- The added check for BOS, EOS, and UNK tokens is critical; ensure that the condition correctly reflects the expected bounds for modelProto.Pieces to prevent runtime errors.
if (modelProto.TrainerSpec.BosId >= modelProto.Pieces.Count ||
src/Microsoft.ML.Tokenizers/Model/SentencePieceOptions.cs:1
- Since the SentencePieceOptions file has been removed, verify that all documentation and external references have been updated accordingly to prevent integration issues.
// Licensed to the .NET Foundation under one or more agreements.
tarekgh
commented
Mar 19, 2025
MemberAuthor
CC @ericstj |
michaelgsharp
approved these changes
Mar 19, 2025
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@## main #7427 +/- ##
==========================================
- Coverage 69.00% 68.96% -0.04%
==========================================
Files 1483 1482 -1 Lines 274563 274198 -365 Branches 28395 28347 -48 ==========================================
- Hits 189455 189103 -352 - Misses 77672 77681 +9 + Partials 7436 7414 -22
Flags with carried forward coverage won't be shown. Click here to find out more.
🚀 New features to boost your workflow:
|
This was referenced Mar 20, 2025
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 freeto 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.
Removing creating sentence piece tokenizer creation with options as it is not needed at least from now. Also, address some older comment #7409 (review) for checking the special tokens before storing them in the vocabulary reverse array.