nicpopovic / fredo Goto Github PK

As following I select torch-1.8.0-cu111-cp37-cp37m-linux_x86_64.whl and my virtual environment is python3.7,but when run train.py I got "RuntimeError: CUDA error: CUBLAS_STATUS_EXECUTION_FAILED when calling `cublasSgemmStridedBatched( handle, opa, opb, m, n, k, &alpha, a, lda, stridea, b, ldb, strideb, &beta, c, ldc, stridec, num_batches)",why I cannot run it?

If you are only interested in the raw data for the tasks, you can directly download sampled episodes here: https://drive.google.com/drive/folders/1PuJSJxqZP4ijxFSBZZ6Fmc0SgR2S8pYU?usp=sharing (test_episodes.zip [~120 MB DL, ~550MB on disk] + train_episodes_single.zip [~330MB DL, ~1.4GB on disk]+ train_episodes_schema.zip [~330MB DL, ~1.4GB on disk])

The two cross-domain test files in the Google Drive (test_cross_domain_1_doc.json and test_cross_domain_3_doc.json) were sampled from the test_docred.json file, which should have been sampled from the test_sicerc.json.

Hello,

Thank you for your work. I tried using the pretrained models you made available on Drive, but am unsure of how to test them with custom data of my own. Is there a script I can use to run the model on my data?

Hello Nicholas & Michael,

Your paper presents an thoughtful benchmark for doc-level RE and I look forward to trying it out. Would be great if you could clarify a few simple questions:

1. Are the support documents for the cross-domain test set (comprising solely of SciERC samples) sampled from your Train+Dev set (62+16 relation types from DocRed)?

2. If the answer to question (1) is yes, is this not zero-shot with the setting that shot is defined as relation types trained/encountered?

3. Just to be certain, where are the support docs for the in-domain test set (16 RT from DocRED) sampled from? (Train+Dev set or in-domain test set)?

Figure illustrating my understanding of the train-test setup in this work: