Binary Classification: How is predicted label computed?

[simonschoe]

Hi there,

I am observing the following (strange) behavior when using pipeline from the transformers library and transformer-interpret:

text = "Now Accord networks is a company in video, and he led the sales team, and the marketing group at Accord, and he took it from start up, sound familiar, it's from start up to $60 million company in two years."
classifier = pipeline('text-classification',  model=model, tokenizer=tokenizer, device=0)
classifier(text)

[{'label': 'LABEL_1', 'score': 0.9711543321609497}]

while transformer-interpret gives me slightly different scores:

explainer = SequenceClassificationExplainer(model, tokenizer)
attributions = explainer(text)
html = explainer.visualize()

In both cases I apply the exact same model and tokenizer...

I am grateful for any hint and/or advice! 🤗