metadata
license: mit
language:
- en
datasets: climate_fever
tags:
- fact-checking
- climate
- text entailment
This model fine-tuned ClimateBert on the textual entailment task using Climate FEVER data. Given (claim, evidence) pairs, the model predicts support (entailment), refute (contradict), or not enough info (neutral). The model has 67% validation accuracy.
from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch
model = AutoModelForSequenceClassification.from_pretrained("amandakonet/climatebert-fact-checking")
tokenizer = AutoTokenizer.from_pretrained("amandakonet/climatebert-fact-checking")
features = tokenizer(['Beginning in 2005, however, polar ice modestly receded for several years'],
['Polar Discovery "Continued Sea Ice Decline in 2005'],
padding='max_length', truncation=True, return_tensors="pt", max_length=512)
model.eval()
with torch.no_grad():
scores = model(**features).logits
label_mapping = ['entailment', 'contradiction', 'neutral']
labels = [label_mapping[score_max] for score_max in scores.argmax(dim=1)]
print(labels)