Fix dtype of weights in from_pretrained when device_map is set #20602
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
|
The documentation is not available anymore as the PR was closed or merged. |
younesbelkada
approved these changes
Dec 5, 2022
Contributor
There was a problem hiding this comment.
Thanks a lot for fixing and making model loading consistent between device_map=auto (or any) and device_map=None !
Just wondering if you need a special safety checker for safetensors (bear in mind that I am not very knowledgable about safetensors - what happens if old_params.dtype == torch.float16 and is_safetensors==True)
Collaborator
Author
|
There is no more safetensors at this stage, ( |
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Outdated suggestions cannot be applied.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?
As reported in #20390, the dtype of the weights after
from_pretrainedis used for a checkpoint is inconsistent betweendevice_map=Noneordevice_mapset:device_map=None(which usesnn.Module.laod_state_dict) will have the dtype of the model stay the same, even if the checkpoints are in a different dtype (so loading a float16 checkpoint in a float32 model gives a float32 model)device_mapset (which manually sets the parameters) will change the dtype of the model to the dtype of the checkpoint (so loading a float16 checkpoint in a float32 model gives a float16 model).This PR addresses this.