Im using the sample code given by huggingface to calculate BertScore, but it is giving f1 for each sample.
predictions = [“hello world”, “general kenobi”]
references = [“hello world”, “general kenobi”]
results = bertscore.compute(predictions=predictions, references=references, model_type=“distilbert-base-uncased”)
print(results)
{‘precision’: [0.9999995827674866, 1.000000238418579], ‘recall’: [0.9999995827674866, 1.000000238418579], ‘f1’: [0.9999995827674866, 1.000000238418579], ‘hashcode’: ‘distilbert-base-uncased_L5_no-idf_version=0.3.12(hug_trans=4.25.1)’}
Should we take the average f1 score to get overall score?