Model Card for Model ID
Based on bert-base-uncased . This model takes in news summary/news headlines/news article and classifies into one of 40 categories (listed below) . Dataset used to traing this model is from Kaggle called News Category Dataset porvided by rishabhmisra.github.io/publications (Please cite this datatset when using this model). Contact us for more info: support@cloudsummary.com.
Below are the output labels:
- arts = 0, arts & culture =1, black voices = 2, business = 3, college = 4, comedy = 5, crime = 6, culture & arts = 7, education = 8, entertainment = 9,environment = 10 fifty=11, food & drink = 12 ,good news = 13, green = 14, healthy living = 15, home & living = 16, impact = 17, latino voices = 18 , media = 19, money = 20 , parenting = 21 , parents = 22 politics = 23, queer voices = 24, religion = 25, science = 26, sports = 27, style = 28, style & beauty = 29 ,taste = 30 ,tech = 31, the worldpost = 32,travel = 33 u.s. news = 34, weddings = 35, weird news = 36, wellness = 37, women = 38 , world news = 39 , worldpost = 40
Model Details
Model Description
- Developed by: CloudSummary (support@cloudsummary.com)
- Model type: Language model
- Language(s) (NLP): English
- License: apache-2.0
- Finetuned from model : bert-base-uncased
Model Sources
Please refer bert-base-uncased for Model Sources.
How to Get Started with the Model
Use the code below to get started with the model.
from transformers import BertTokenizer, BertForSequenceClassification
import torch
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
model = BertForSequenceClassification.from_pretrained ("cssupport/bert-news-class").to(device)
def predict(text):
id_to_class = {0: 'arts', 1: 'arts & culture', 2: 'black voices', 3: 'business', 4: 'college', 5: 'comedy', 6: 'crime', 7: 'culture & arts', 8: 'education', 9: 'entertainment', 10: 'environment', 11: 'fifty', 12: 'food & drink', 13: 'good news', 14: 'green', 15: 'healthy living', 16: 'home & living', 17: 'impact', 18: 'latino voices', 19: 'media', 20: 'money', 21: 'parenting', 22: 'parents', 23: 'politics', 24: 'queer voices', 25: 'religion', 26: 'science', 27: 'sports', 28: 'style', 29: 'style & beauty', 30: 'taste', 31: 'tech', 32: 'the worldpost', 33: 'travel', 34: 'u.s. news', 35: 'weddings', 36: 'weird news', 37: 'wellness', 38: 'women', 39: 'world news', 40: 'worldpost'}
# Tokenize the input text
inputs = tokenizer(text, return_tensors='pt', truncation=True, max_length=512, padding='max_length').to(device)
with torch.no_grad():
logits = model(inputs['input_ids'], inputs['attention_mask'])[0]
# Get the predicted class index
pred_class_idx = torch.argmax(logits, dim=1).item()
return id_to_class[pred_class_idx]
text ="The UK’s growing debt burden puts it on shaky ground ahead of upcoming assessments by the three main credit ratings agencies. A downgrade to its credit rating, which is a reflection of a country’s creditworthiness, could raise borrowing costs further still, although the impact may be limited."
predicted_class = predict(text)
print(predicted_class)
#OUTPUT : business
Uses
[More Information Needed]
Direct Use
Could used in application where natural language is to be converted into SQL queries. [More Information Needed]
Out-of-Scope Use
[More Information Needed]
Bias, Risks, and Limitations
[More Information Needed]
Recommendations
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
Technical Specifications
Model Architecture and Objective
Compute Infrastructure
Hardware
one P6000 GPU
Software
Pytorch and HuggingFace
Citation
Misra, Rishabh. "News Category Dataset." arXiv preprint arXiv:2209.11429 (2022). Misra, Rishabh and Jigyasa Grover. "Sculpting Data for ML: The first act of Machine Learning." ISBN 9798585463570 (2021). Tandon, Karan. "This LLM is based on BERT (2018) a bidirectional Transformer. cssupport/bert-news-class was finetuned using AdamW with the help of NVIDIA AMP and trained in 45 minutes on one P6000 GPU. This model accepts news summary/news headlines/news article and classifies into one of 40 categories"