一个经过训练的交叉注意力NLI模型，用于零-shot和少-shot文本分类。

基础模型是 xlm-roberta-base ，使用了 here 的代码进行训练；在 SNLI 、 MNLI 、 ANLI 和 XNLI 上进行训练。

用法：

from transformers import AutoModelForSequenceClassification, AutoTokenizer
import torch
import numpy as np

model = AutoModelForSequenceClassification.from_pretrained("symanto/xlm-roberta-base-snli-mnli-anli-xnli")
tokenizer = AutoTokenizer.from_pretrained("symanto/xlm-roberta-base-snli-mnli-anli-xnli")

input_pairs = [
               ("I like this pizza.", "The sentence is positive."),
               ("I like this pizza.", "The sentence is negative."),
               ("I mag diese Pizza.", "Der Satz ist positiv."),
               ("I mag diese Pizza.", "Der Satz ist negativ."),
               ("Me gusta esta pizza.", "Esta frase es positivo."),
               ("Me gusta esta pizza.", "Esta frase es negativo."),
]
inputs = tokenizer(input_pairs, truncation="only_first", return_tensors="pt", padding=True)
logits = model(**inputs).logits
probs = torch.softmax(logits, dim=1)
probs = probs[..., [0]].tolist()
print("probs", probs)
np.testing.assert_almost_equal(probs, [[0.83], [0.04], [1.00], [0.00], [1.00], [0.00]], decimal=2)