”wrap_with_autocast“ is not implemented

[Jiangggg1995]

Hello,

I am currently working on translating the Llama2 model (from transformers==4.40.2 and pytorch==2.5.1) to MLIR using torch-mlir. However, I encountered the following error:

NotImplementedError: Higher-order operation 'wrap_with_autocast' not implemented in the FxImporter (tried '_import_hop_wrap_with_autocast')
It appears that there is an unimplemented operation, wrap_with_autocast, which is found within the LlamaRotaryEmbedding class. This operation seems to be generated by the context manager with torch.autocast(device_type=device_type, enabled=False).

I would appreciate any guidance or assistance on how to resolve this problem.

Thank you!