⚠️ [CLAP] Fix dtype of logit scales in init #25682
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
sanchit-gandhi
commented
Aug 23, 2023
| @@ -1956,8 +1955,8 @@ def __init__(self, config: ClapConfig): | |||
| text_config = config.text_config | |||
| audio_config = config.audio_config | |||
|
|
|||
| self.logit_scale_a = nn.Parameter(torch.tensor(np.log(config.logit_scale_init_value))) | |||
There was a problem hiding this comment.
The aforementioned behaviour is a result of the np.log operation defaulting to float64
There was a problem hiding this comment.
Given the original code we might need to init in float64 then cast to float if it makes a difference. No idea if the actual value save is in float64!
There was a problem hiding this comment.
The parameters are initialised in float64 but are stored in float32 in the state dict
ArthurZucker
approved these changes
Aug 23, 2023
|
The documentation is not available anymore as the PR was closed or merged. |
|
Note that in the original repo, the model is always cast to float16 for all training / inference. Thus, they likely never used the model in it's default dtype, and always relied on explicitly casting to float16 |
sanchit-gandhi
changed the title
parambharat
pushed a commit
to parambharat/transformers
that referenced
this pull request
Sep 26, 2023
[CLAP] Fix dtype of logit scales
blbadger
pushed a commit
to blbadger/transformers
that referenced
this pull request
Nov 8, 2023
[CLAP] Fix dtype of logit scales
EduardoPach
pushed a commit
to EduardoPach/transformers
that referenced
this pull request
Nov 18, 2023
[CLAP] Fix dtype of logit scales
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Outdated suggestions cannot be applied.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?
The dtype of the CLAP logit scale parameters was always float64 by default (even if the rest of the model was initialised in float32). This PR fixes the logit scales, such that they respect the default dtype of the model.