test_quip.py

from transformers import AutoConfig, AutoModelForSequenceClassification, AutoTokenizer

device = "cuda"
predictor = AutoModelForSequenceClassification.from_pretrained("/home/afabbri/evaluation/qafacteval/models/quip-512-mocha").to(device)
predictor.eval()

tokenizer = AutoTokenizer.from_pretrained("/home/afabbri/evaluation/qafacteval/models/quip-512-mocha")
batch_size = 32


context = "The Yankees beat the Braves in the World Series."


questions = ["Who won the World Series?", "Who beat the Braves?"]
answers = ["The Yankees", "The Yankees"]
predictions = ["The Yankees", "The Yankees"]

batch_sentences = []
for question, answer, prediction in zip(questions, answers, predictions):
    sentence1 = f"{question} <q> {answer} <r> {prediction} <c> {context}"
    batch_sentences.append(sentence1)

inputs = tokenizer(batch_sentences, max_length=512,  truncation=True, padding="max_length", return_tensors="pt")
import pdb;pdb.set_trace()
outputs = predictor(input_ids=inputs["input_ids"].to(device), attention_mask=inputs["attention_mask"].to(device))
outputs = [x[0] for x in outputs[0].cpu().tolist()]
import pdb;pdb.set_trace()
print("HI")