Closed EdwardZH closed 4 years ago
Are you using our provided script or are you trying to load the model yourself?
Sure, I use the provided scripts. Only the rationale selection part with RoBERTa lager does not work, and then I replace it to sciBERT it works.
I have double-checked that the script runs fine for RoBERTa Large on my side. What GPU are you using? Maybe the GPU runs out of VRAM for RoBERTa Large model? If there are more issues I will need to know what's the exact error message it produces.
"Unable to load weights from pytorch checkpoint file. "
OSError: Unable to load weights from pytorch checkpoint file. If you tried to load a PyTorch model from a TF 2.0 checkpoint, please set from_tf=True.
terminate called after throwing an instance of 'c10::Error'
what(): owning_ptr == NullType::singleton() || owningptr->refcount.load() > 0 INTERNAL ASSERT FAILED at /pytorch/c10/util/intrusive_ptr.h:348, please report a bug to PyTorch. intrusive_ptr: Can only intrusive_ptr::reclaim() owning pointers that were created using intrusive_ptr::release(). (reclaim at /pytorch/c10/util/intrusive_ptr.h:348)
frame #0: c10::Error::Error(c10::SourceLocation, std::string const&) + 0x33 (0x7f846cc53193 in /home/liuzhenghao/.conda/envs/fever/lib/python3.7/site-packages/torch/lib/libc10.so)
frame #1:
I have no idea why would you have that error.
I created this temporary Colab notebook to show how it is working on my side. https://colab.research.google.com/drive/1QXS901zB65pGO5cF4QvDs8O1fH4x10Kk?usp=sharing
I download checkpoint again. It works. Thank you for your help.
I follow the pipeline and found errors when I use rationale_roberta_large_scifact. I think the checkpoint is not pytorch form and miss vocab file.