ALBERT for Math AR

This model is further pre-trained on the Mathematics StackExchange questions and answers. It is based on Albert base v2 and uses the same tokenizer. In addition to pre-training the model was finetuned on Math Question Answer Retrieval. The sequence classification head is trained to output a relevance score if you input the question as the first segment and the answer as the second segment. You can use the relevance score to rank different answers for retrieval.

Usage

# based on https://huggingface.co/docs/transformers/main/en/task_summary#sequence-classification
from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

tokenizer = AutoTokenizer.from_pretrained("albert-base-v2")
model = AutoModelForSequenceClassification.from_pretrained("AnReu/albert-for-math-ar-base-ft")

classes = ["non relevant", "relevant"]

sequence_0 = "How can I calculate x in $3x = 5$"
sequence_1 = "Just divide by 3: $x = \\frac{5}{3}$"
sequence_2 = "The general rule for squaring a sum is $(a+b)^2=a^2+2ab+b^2$"

# The tokenizer will automatically add any model specific separators (i.e. <CLS> and <SEP>) and tokens to
# the sequence, as well as compute the attention masks.
irrelevant = tokenizer(sequence_0, sequence_2, return_tensors="pt")
relevant = tokenizer(sequence_0, sequence_1, return_tensors="pt")

irrelevant_classification_logits = model(**irrelevant).logits
relevant_classification_logits = model(**relevant).logits

irrelevant_results = torch.softmax(irrelevant_classification_logits, dim=1).tolist()[0]
relevant_results = torch.softmax(relevant_classification_logits, dim=1).tolist()[0]

# Should be irrelevant
for i in range(len(classes)):
    print(f"{classes[i]}: {int(round(irrelevant_results[i] * 100))}%")

# Should be relevant
for i in range(len(classes)):
    print(f"{classes[i]}: {int(round(relevant_results[i] * 100))}%")

Reference

If you use this model, please consider referencing our paper:

@inproceedings{reusch2021tu_dbs,
  title={TU\_DBS in the ARQMath Lab 2021, CLEF},
  author={Reusch, Anja and Thiele, Maik and Lehner, Wolfgang},
  year={2021},
  organization={CLEF}
}