UX4567/Speak-English-AI-3B-v1.0

Model Card for Speak-English-AI-3B-v1.0

A lightweight Hindi <-> English translator for spoken English practice, made by Kartik Sharma from Jaipur.

Model Details

Model Description

This is a LoRA fine-tuned version of NLLB 600M. It fixes the common error where other models translate Jaipur as Plu or Cartozy. This model correctly translates daily conversation sentences.

• —Developed by: Kartik Sharma
• —Shared by: Kartik Sharma from Jaipur
• —Model type: Seq2Seq Translation (NLLB + LoRA)
• —Language(s) (NLP): Hindi (hinDeva), English (engLatn)
• —License: mit
• —Finetuned from model: facebook/nllb-200-distilled-600M

Uses

Direct Use

Anyone who wants to practice Hindi to English speaking. Example: मैं बाजार जा रहा हूँ -> I am going to market

Downstream Use

Can be plugged into a Gradio App, WhatsApp bot, or English learning app.

How to Get Started with the Model

Use the code below to get started with the model.

from peft import PeftModel
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

!pip install -U torchao

adapter_id = "UX4567/Speak-English-AI-3B-v1.0"

tokenizer = AutoTokenizer.from_pretrained(adapter_id)
model = AutoModelForSeq2SeqLM.from_pretrained(adapter_id)


def translate(text, src_lang, tgt_lang):
    tokenizer.src_lang = src_lang
    inputs = tokenizer(text, return_tensors="pt")
    out = model.generate(
        **inputs,
        max_new_tokens=100
    )
    return tokenizer.batch_decode(out, skip_special_tokens=True)[0]

# Hindi -> English
print(translate("मैं बाजार जा रहा हूँ", "hin_Deva", "eng_Latn"))
# English -> Hindi
print(translate("I am from Jaipur", "eng_Latn", "hin_Deva"))