itsmeaminul/bangla-punct-model

🔠 Bangla Punctuation Restoration Model

This is a token classification model fine-tuned to restore punctuation in Bangla sentences. It predicts and inserts punctuation marks such as comma (,), period (।), question mark (?), exclamation (!), and more based on context.

---

🧠 Model Details

• Model Base: sagorsarker/bangla-bert-base
• Architecture: BERT-based token classification
• Task: Punctuation Restoration for Bangla
• Trained on: Custom corpus of Bangla unpunctuated sentences aligned with punctuated references
• Labels:
  • O – No punctuation
  • COMMA – ,
  • PERIOD – ।
  • QUESTION – ?
  • EXCLAMATION – !
  • SEMICOLON – ;
  • COLON – :
  • HYPHEN – -

---

📈 Evaluation Metrics

Metric	Value
F1-score (macro)	70.4%
Sentence accuracy	71.8%

Note: Metrics are computed on a held-out Bangla test set.

---

🚀 Inference Example

Input (Unpunctuated): আমি তোমাকে বলেছিলাম তুমি কেন আসোনি

Output (Punctuated) : আমি তোমাকে বলেছিলাম, তুমি কেন আসোনি?