mbart_ru_sum_gazeta

Introduction

The MBART_RU_SUM_GAZETA is a model tailored for summarizing Russian news articles, particularly from Gazeta.ru. It is built using the mBART architecture and offers capabilities for text-to-text generation.

Architecture

The model is a ported version of a fairseq model specifically designed for summarization tasks. It utilizes the mBART architecture, which is well-suited for multilingual text processing.

Training

Training Data

• Dataset: The model was trained on the Gazeta dataset, a collection of Russian news articles.

Training Procedure

• Script: Utilized a Fairseq training script available in the repository.
• Porting: Conducted porting using a Colab notebook, making it accessible for further experimentation and deployment.

Evaluation

The model was evaluated using metrics like ROUGE (R-1-f, R-2-f, R-L-f), chrF, METEOR, and BLEU. The evaluation demonstrated competitive performance, particularly on Gazeta.ru articles.

Guide: Running Locally

1. Setup:

   • Install the transformers library from Hugging Face.
   • Install PyTorch if not already done.

2. Code Example:

   from transformers import MBartTokenizer, MBartForConditionalGeneration
   
   model_name = "IlyaGusev/mbart_ru_sum_gazeta"
   tokenizer = MBartTokenizer.from_pretrained(model_name)
   model = MBartForConditionalGeneration.from_pretrained(model_name)
   
   article_text = "..."
   
   input_ids = tokenizer(
       [article_text],
       max_length=600,
       padding="max_length",
       truncation=True,
       return_tensors="pt"
   )["input_ids"]
   
   output_ids = model.generate(
       input_ids=input_ids,
       no_repeat_ngram_size=4
   )[0]
   
   summary = tokenizer.decode(output_ids, skip_special_tokens=True)
   print(summary)
   

3. Hardware Recommendation:

   • For efficient processing, using cloud GPUs such as those from AWS, GCP, or Azure is recommended.

License

The MBART_RU_SUM_GAZETA model is released under the Apache-2.0 license, allowing for both personal and commercial use.