Add MACE-OFF23 models #275
Conversation
Fix selection of device to allow integer
Update version to 0.3.2 for conda installation
Update version for conda release
Update to 0.3.2 to overwrite previous tag
Add small model and warning for float32 accuracy in mace_mp
update the small model to energy model
|
@davkovacs I think we should also add a paragraph in the readme about the models |
| def mace_off( | ||
| model: Union[str, Path] = None, | ||
| device: str = "", | ||
| default_dtype: str = "float32", |
There was a problem hiding this comment.
The default dtype does not match the doc string.
There was a problem hiding this comment.
good catch, I am updating it.
| "Model download failed" | ||
| ) from exc | ||
|
|
||
| device = device or ("cuda" if torch.cuda.is_available() else "cpu") |
There was a problem hiding this comment.
maybe return the raw model here as the print statements are not relevant for users that just want the model.
|
Ready to be merged. |
|
Might be good to add a test as well to make sure nothing gets broken in the future? |
|
You can use the same test as the mace_mp |
|
Can you test the tests please! @jthorton |
|
ready to merge? |
|
@ilyes319 if you are also happy with the tests and everything else, I think it is ready. |
| msg = f"Using MACE-OFF23 MODEL for MACECalculator with {model}" | ||
| print(msg) |
There was a problem hiding this comment.
Looks like we missed this print which should only trigger if we don't just want to return the model.
This PR implements the interface to the MACE-OFF23 models.