Charlelie Laurent

Enable saving/loading Module compiled with torch dynamo and dynamically set AMP in CorrDiff Module wrappers

1

# PhysicsNeMo Pull Request ## Description This PR: - Implements utilities to strip out torch dynamo wrappers when attempting to save a `Module` that was compiled with `model = torch.compile(model)`....

### Is this a new feature, an improvement, or a change to existing functionality? New Feature ### How would you describe the priority of this feature request Low (would be...

# PhysicsNeMo Pull Request ## Description This PR improves the docstring for the method `Module.from_checkpoint` with added details regarding the mechanism to infer the class, as well as recommended usage....

### How would you describe the priority of this documentation request Critical (currently preventing usage) ### Is this for new documentation, or an update to existing docs? New ### Describe...

### Version 1.2.0 ### On which installation method(s) does this occur? _No response_ ### Describe the issue ## Problem @nbren reported that the [UNet example](https://docs.nvidia.com/physicsnemo/latest/physicsnemo/api/models/modules.html#how-to-write-your-own-physicsnemo-model) in the documentation is not...

### How would you describe the priority of this documentation request Critical (currently preventing usage) ### Is this for new documentation, or an update to existing docs? New ### Describe...

### Is this a new feature, an improvement, or a change to existing functionality? New Feature ### How would you describe the priority of this feature request Critical (currently preventing...

### Is this a new feature, an improvement, or a change to existing functionality? New Feature ### How would you describe the priority of this feature request Critical (currently preventing...

# PhysicsNeMo Pull Request ## Description **Note:** this PR is currently a draft and is open for feedback and suggestions. It will not be merged in its current form and...

Added backend context manager to select SDPA implementation

1

# PhysicsNeMo Pull Request ## Description ### Context PR #954 changed the computation of the `Attention` forward defined in `physicsnemo/models/diffusion/layers.py` such that it is now based on `torch.nn.scaled_dot_product_attention` (instead of...