Using HuggingFace ViT implementation (#219) #303
Conversation
| @@ -119,6 +144,7 @@ def __init__(self, **kwargs: Any) -> None: | |||
| num_heads=12, | |||
| hidden_dim=768, | |||
| mlp_dim=3072, | |||
| pretrained_name="google/vit-base-patch16-224", | |||
There was a problem hiding this comment.
why still pass the other arguments? they aren't used, right? or is this a requirement for reloading the model?
There was a problem hiding this comment.
Ok, I thought the idea was to reload it without pretrained_name. Now, it will not be able to reloaded since pretrained_name is missing? Can we maybe add a test for this model or do we already have that?
There was a problem hiding this comment.
model_fn loads through from_state_dict if model_state_url is specified. So reloading (in principle) should be able to restore the weights.
There was a problem hiding this comment.
What would a test for this do?
| prediction_strategy: Optional[PredictionStrategy] = None, | ||
| add_icarl_class_means: bool = True, | ||
| pretrained_name: Optional[str] = None, |
There was a problem hiding this comment.
maybe make this the first argument?
Coverage reportThe coverage rate went from
Diff Coverage details (click to unfold)src/renate/benchmark/models/vision_transformer.py
|
Solves #219
By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.