push_to_hub is broken due to some tensors sharing memory #48
Description
Currently, when saving a model and its processor after training, the following error is thrown:
...
RuntimeError:
Some tensors share memory, this will lead to duplicate memory on disk and potential differences when loading them again: [{'language_model.lm_head.weight', 'language_model.model.embed_tokens.weight'}].
A potential way to correctly save your model is to use `save_model`.
More information at https://huggingface.co/docs/safetensors/torch_shared_tensorsThis error indicated that 2 layers tensors are sharing memory and this issue prevents it from saving it correctly.