Add dtype arg to Gemma HF conversion script #1452
Conversation
| def _set_default_tensor_type(dtype: torch.dtype): | ||
| """Sets the default torch dtype to the given dtype.""" | ||
| torch.set_default_dtype(dtype) | ||
| yield |
There was a problem hiding this comment.
The yield allows whatever is in the context (in this case, the checkpoint conversion) to fully execute. After that is done, this will reset to the default torch dtype.
There was a problem hiding this comment.
Ah i see. in that case, do we need to care?
No real problem with this, but overall is intended to be running as a toplevel script right? Is there some scenario in which more python would be changed after the conversion inside the same python execution?
There was a problem hiding this comment.
I think it's good practice to keep the dtype change within the context of the scope rather than affecting it globally. If we add any more code outside the scope after, we'd need to be aware of a global property change.
No description provided.