fix: refactor post_processor logic and add test #2137
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
Narsil
reviewed
Jun 27, 2024
router/src/main.rs
Outdated
| if let Some(class) = &tokenizer_config.tokenizer_class { | ||
| if class == "LlamaTokenizer" || class == "LlamaTokenizerFast" { | ||
| tracing::info!("Overriding LlamaTokenizer with TemplateProcessing to follow python override defined in https://github.com/huggingface/transformers/blob/4aa17d00690b7f82c95bb2949ea57e22c35b4336/src/transformers/models/llama/tokenization_llama_fast.py#L203-L205"); | ||
| if let Some(post_processor) = create_post_processor(tokenizer, &tokenizer_config) { |
There was a problem hiding this comment.
We need to override only is the current post_processor is None, forgot that condition.
There was a problem hiding this comment.
updated and added in latest commit
router/src/main.rs
Outdated
| pub fn create_post_processor( | ||
| tokenizer: &Tokenizer, | ||
| tokenizer_config: &HubTokenizerConfig, | ||
| ) -> Option<TemplateProcessing> { |
There was a problem hiding this comment.
Why not return an Result and remove all those unwrap() into ??
There was a problem hiding this comment.
great point! updated in latest
| if let Some(class) = &tokenizer_config.tokenizer_class { | ||
| if class == "LlamaTokenizer" || class == "LlamaTokenizerFast" { | ||
| tracing::info!("Overriding LlamaTokenizer with TemplateProcessing to follow python override defined in https://github.com/huggingface/transformers/blob/4aa17d00690b7f82c95bb2949ea57e22c35b4336/src/transformers/models/llama/tokenization_llama_fast.py#L203-L205"); | ||
| if let Some(post_processor) = create_post_processor(tokenizer, &tokenizer_config) { | ||
| tokenizer.with_post_processor(post_processor); |
There was a problem hiding this comment.
Let's log only when we actually override, here we would log even if the function fails.
There was a problem hiding this comment.
updated and moved after create_post_processor succeeds
Narsil
reviewed
Jun 27, 2024
|
|
||
| let mut single = String::new(); | ||
| let mut pair = String::new(); | ||
| let mut special_tokens = Vec::new(); |
There was a problem hiding this comment.
Can we keep the actual vec ? String is really a clutch.
glegendre01
pushed a commit
that referenced
this pull request
Jul 2, 2024
* fix: refactor post_processor logic and add test * fix: remove dev comment * fix: adjust when post_processor is overridden and improve create_post_processor
yuanwu2017
pushed a commit
to yuanwu2017/tgi-gaudi
that referenced
this pull request
Sep 26, 2024
* fix: refactor post_processor logic and add test * fix: remove dev comment * fix: adjust when post_processor is overridden and improve create_post_processor
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.
This PR updates the logic for adding a
post_processorto a tokenizer and should resolve the issues in CI https://github.com/huggingface/text-generation-inference/actions/runs/9698059633/job/26765404563This PR aims to match the same functionality in
transformers/models/llama/tokenization_llama_fast.pyand fixes an issue where the single and pairs were not correctly constructed and applied. This resulted in a missingbos_tokenand subsequently an issue with accessing slots inbatch.slots[batch.slot_indices]. Also a test is added for that specific case