metadata

language:
  - fa
tags:
  - Wikipedia
  - Summarizer
  - bert2bert
task_categories:
  - summarization
  - text generation
task_ids:
  - news-articles-summarization
license:
  - apache-2.0
multilinguality:
  - monolingual
datasets:
  - pn-summary
  - XL-Sum
metrics:
  - rouge-1
  - rouge-2
  - rouge-l

WikiBert2WikiBert

Bert language models can be employed for Summarization tasks. WikiBert2WikiBert is an encoder-decoder transformer model that is initialized using the Persian WikiBert Model weights. The WikiBert Model is a Bert language model which is fine-tuned on Persian Wikipedia. After using the WikiBert weights for initialization, the model is trained for five epochs on PN-summary and Persian BBC datasets.

How to Use:

You can use the code below to get the model's outputs, or you can simply use the demo on the right.

from transformers import (
    BertTokenizerFast,
    EncoderDecoderConfig,
    EncoderDecoderModel,
    BertConfig
)

model_name = 'Arashasg/WikiBert2WikiBert'
tokenizer = BertTokenizerFast.from_pretrained(model_name)
config = EncoderDecoderConfig.from_pretrained(model_name)
model = EncoderDecoderModel.from_pretrained(model_name, config=config)


def generate_summary(text):
    inputs = tokenizer(text, padding="max_length", truncation=True, max_length=512, return_tensors="pt")
    input_ids = inputs.input_ids.to("cuda")
    attention_mask = inputs.attention_mask.to("cuda")

    outputs = model.generate(input_ids, attention_mask=attention_mask)

    output_str = tokenizer.batch_decode(outputs, skip_special_tokens=True)


    return output_str

input = 'your input comes here'
summary = generate_summary(input)

Evaluation

I separated 5 percent of the pn-summary for evaluation of the model. The rouge scores of the model are as follows:

Rouge-1	Rouge-2	Rouge-l
38.97%	18.42%	34.50%