Hello, I am John who is currently contributing to Issue 17935 (TF: XLA generation not working properly in some models · Issue #17935 · huggingface/transformers · GitHub) with @AdiaWu and @katiele47. We are trying to figure out the root of the software defects existing in the OPT model, but it is really hard to make a progress. Specifically, we are feeling difficulties what is preventing from generating XLA in OPT model.
Attached screenshots show the simulations in a local environmnet. Would you mind giving me any suggestions?
