ayhay/BanglaText2Gloss

Model Card for Model ID

Translate Bangla sentences to Bangla Sign Language(BdSL) gloss with a lightweight model!

Model Details

Model Description

• Developed by: Sharif Mohammad Abdullah
• Language(s) (NLP): Bangla to Bangla Gloss
• Finetuned from model: mbart-large-50

Model Sources [optional]

• Paper [optional]: https://arxiv.org/abs/2504.02293

Uses

You can use the following snippet to do a test run for a sample sentence:

from transformers import MBartForConditionalGeneration, AutoTokenizer
from peft import PeftModel

model_path = "ayhay/BanglaText2Gloss" 
base_id = "facebook/mbart-large-50"

# 1. Load Tokenizer & Model with Adapters
tokenizer = AutoTokenizer.from_pretrained(model_path)
base_model = MBartForConditionalGeneration.from_pretrained(base_id)
model = PeftModel.from_pretrained(base_model, model_path)

# 2. Run Translation
text = "আপনি কেমন আছেন?" 
inputs = tokenizer(text, return_tensors="pt")

output_tokens = model.generate(**inputs, max_new_tokens=50)
gloss = tokenizer.batch_decode(output_tokens, skip_special_tokens=True)

print(f"Bangla: {text}")
print(f"Sign Gloss: {gloss[0]}")

Out-of-Scope Use

The model will not work for any languages other than Bangla.

-->

Citation [optional]

If you use this model in your works, pleasae cite using the following bibtex format:

BibTeX:

[More Information Needed]

Framework versions

• PEFT 0.18.0