README.md

metadata

language: bn
tags:
  - collaborative
  - bengali
  - SequenceClassification
license: apache-2.0
datasets: IndicGlue
metrics:
  - Loss
  - Accuracy
  - Precision
  - Recall

sahajBERT News Article Classification

Model description

sahajBERT fine-tuned for news article classification using the sna.bn split of IndicGlue.

The model is trained for classifying articles into 5 different classes:

Label id	Label
0	kolkata
1	state
2	national
3	sports
4	entertainment
5	international

Intended uses & limitations

How to use

You can use this model directly with a pipeline for Sequence Classification:

from transformers import AlbertForSequenceClassification, TextClassificationPipeline, PreTrainedTokenizerFast

# Initialize tokenizer
tokenizer = PreTrainedTokenizerFast.from_pretrained("neuropark/sahajBERT-NCC")

# Initialize model
model = AlbertForSequenceClassification.from_pretrained("neuropark/sahajBERT-NCC")

# Initialize pipeline
pipeline = TextClassificationPipeline(tokenizer=tokenizer, model=model)

raw_text = "এই ইউনিয়নে ৩ টি মৌজা ও ১০ টি গ্রাম আছে ।" # Change me
output = pipeline(raw_text)

Limitations and bias

WIP

Training data

The model was initialized with pre-trained weights of sahajBERT at step 18149 and trained on the sna.bn split of IndicGlue.

Training procedure

Coming soon!

Eval results

accuracy: 0.920623671155209

loss: 0.2719293534755707

macro_f1: 0.8924089161713425

macro_precision: 0.891858452957785

macro_recall: 0.8978917764271065

micro_f1: 0.920623671155209

micro_precision: 0.920623671155209

micro_recall: 0.920623671155209

weighted_f1: 0.9205158122362266

weighted_precision: 0.9236142214371135

weighted_recall: 0.920623671155209

BibTeX entry and citation info

Coming soon!