Granite speech - minor fixes to support training with the HF trainer #38833
+9
−24
Conversation
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
avoid unused parameters that DDP does not like
trainers often pass this argument automatically
this ensures save_pretrained will not crash when saving the processor during training https://github.com/huggingface/transformers/blob/d5d007a1a0f0c11a726a54c8f00bd71825f84d02/src/transformers/feature_extraction_utils.py#L595
| @@ -50,15 +50,16 @@ def __init__( | |||
| **kwargs, | |||
| ): | |||
| super().__init__(**kwargs) | |||
| self.sampling_rate = sampling_rate | |||
There was a problem hiding this comment.
I think this isn't used currently
There was a problem hiding this comment.
Right, it's not used. I added it to stay consistent with other audio feature extractors that have this property.
avihu111
commented
Jun 16, 2025
| @@ -88,7 +88,9 @@ def __call__( | |||
| else: | |||
| audio_inputs = {} | |||
|
|
|||
| text_inputs = self.tokenizer(prompt_strings, padding=True, **kwargs) | |||
| if "padding" not in kwargs: | |||
There was a problem hiding this comment.
avoids a crash when trainers pass padding=True to the processor
| @@ -92,7 +92,7 @@ def forward(self, hidden_states: torch.Tensor) -> torch.Tensor: | |||
| hidden_states = hidden_states.view(batch_size * nblocks, self.window_size, dim) | |||
|
|
|||
| query_output = self.qformer( | |||
| query_embeds=self.query.data, | |||
| query_embeds=self.query, | |||
There was a problem hiding this comment.
Bugfix. When using .data this trainable parameter did not receive gradients.
| # Currently lazily initialized | ||
| self.melspec = None | ||
| requires_backends(self, ["torchaudio"]) | ||
| self.mel_filters = torchaudio.transforms.MelSpectrogram(**self.melspec_kwargs) |
There was a problem hiding this comment.
removed the lazy init, and renamed it to mel_filters. This specific name avoids a crash when serializing the 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.
What does this PR do?
Minor updates to
granite_speechto enable finetuning it with HF trainers.padding=Trueto the processor.datafrom a forward callmelspectomel_filtersto leverage this, which avoids a crash onsave_pretrainedBefore submitting
Pull Request section?
to it if that's the case.
documentation guidelines, and
here are tips on formatting docstrings.
Who can review?
Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.
CC: @ArthurZucker @eustlb @alex-jw-brooks @avishaiElmakies @gsaon