Huggingface Model: BART-MNLI-ZeroShot-Text-Classification
This is a Huggingface model fine-tuned on the CNN news dataset for zero-shot text classification task using BART-MNLI. The model achieved an f1 score of 94% and an accuracy of 94% on the CNN test dataset with a maximum length of 128 tokens.
Authors
This work was done by CHERGUELAINE Ayoub & BOUBEKRI Faycal
Original Model
Model Architecture
The BART-Large-MNLI model has 12 transformer layers, a hidden size of 1024, and 406 million parameters. It is pre-trained on the English Wikipedia and BookCorpus datasets, and fine-tuned on the Multi-Genre Natural Language Inference (MNLI) task.
Dataset
The CNN news dataset was used for fine-tuning the model. This dataset contains news articles from the CNN website and is labeled into 6 categories, including politics, health, entertainment, tech, travel, world, and sports.
Fine-tuning Parameters
The model was fine-tuned for 1 epoch on a maximum length of 256 tokens. The training took approximately 6 hours to complete.
Evaluation Metrics
The model achieved an f1 score of 94% and an accuracy of 94% on the CNN test dataset with a maximum length of 128 tokens.
Usage
The model can be used for zero-shot text classification tasks on news articles. It can be accessed via the Huggingface Transformers library using the following code:
from transformers import pipeline, AutoTokenizer, AutoModelForSequenceClassification
tokenizer = AutoTokenizer.from_pretrained("AyoubChLin/Bart-MNLI-CNN_news")
model = AutoModelForSequenceClassification.from_pretrained("AyoubChLin/Bart-MNLI-CNN_news")
classifier = pipeline(
"zero-shot-classification",
model=model,
tokenizer=tokenizer,
device=0
)
Acknowledgments
We would like to acknowledge the Huggingface team for their open-source implementation of transformer models and the CNN news dataset for providing the labeled dataset for fine-tuning.