Open 1024er opened 2 years ago
You can construct dev data by using the downstream retrieval tasks in KILT or splitting a portion of the data from the training data.
hello, It's a impressive work. I've got some confusion about evaluation process. I found that evaluation on test data was provided, but the test data didn't have any gold answers . Can you tell me how to evaluate on the test datasets?
In this procedure ``bash scripts/preprocess_fairseq.sh $DATASET_PATH $MODEL_PATH''
FileNotFoundError: [Errno 2] No such file or directory: 'data/knowledge//dev.source'