Model Card for English to Hinglish Translation Model
Model Details
Model Description
This is a fine-tuned T5-small model for translating English sentences into Hinglish (a mix of Hindi and English written in Latin script). The model was trained using LoRA (Low-Rank Adaptation) to optimize training efficiency.
- Developed by: Team AI-Pradarshan(Rashmi Rai, Ayesha, Bitasta)
- Funded by [optional]: [More Information Needed]
- Shared by [optional]: [Your Hugging Face Username]
- Model type: Sequence-to-Sequence Language Model
- Language(s) (NLP): English, Hinglish
- License: MIT
- Finetuned from model [optional]: google-t5/t5-small
Model Sources [optional]
- Repository: https://huggingface.co/rairashmi/hinglish_translation_lora
- Dataset: rairashmi/en-to-hinglish-dataset
Uses
Direct Use
This model can be used to translate English sentences into Hinglish text directly via Hugging Face Transformers.
Downstream Use [optional]
The model can be fine-tuned further or integrated into conversational AI systems and chatbots.
Out-of-Scope Use
- This model is not designed for real-time conversational applications.
- It may not perform well on non-standard or highly domain-specific English text.
Bias, Risks, and Limitations
- The dataset used may contain inherent biases in Hinglish translation styles.
- Accuracy may vary for different dialects and sentence structures.
Recommendations
Users should be aware of translation inconsistencies and verify translations for critical applications.
How to Get Started with the Model
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
model_name = "rairashmi/hinglish_translation_lora"
model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
tokenizer = AutoTokenizer.from_pretrained(model_name)
def translate_english_to_hinglish(text):
inputs = tokenizer(f"translate English to Hinglish: {text}", return_tensors="pt", padding=True, truncation=True)
outputs = model.generate(**inputs)
return tokenizer.decode(outputs[0], skip_special_tokens=True)
sentence = "How are you?"
translation = translate_english_to_hinglish(sentence)
print(f"🔹 English: {sentence}")
print(f"🟢 Hinglish: {translation}")
Training Details
Training Data
The model was trained on the rairashmi/en-to-hinglish-dataset, a parallel corpus of English-Hinglish text pairs.
Training Procedure
Preprocessing [optional]
- Tokenized using the T5 tokenizer
- Padding and truncation applied with a max length of 128
Training Hyperparameters
- Learning Rate: 2e-5
- Batch Size: 8
- Epochs: 2
- Mixed Precision: FP16
Speeds, Sizes, Times [optional]
- Training took approximately X hours on an A100 GPU
- Model size: T5-Small with LoRA adapters
Evaluation
Testing Data, Factors & Metrics
Testing Data
- Evaluated on a held-out validation split of the dataset.
Factors
- Evaluated across different sentence lengths and complexities.
Metrics
- BLEU Score: X.XX (Evaluated using
sacrebleu)
Results
- The model achieves X.XX BLEU Score on the test set.
Model Examination [optional]
[More Information Needed]
Environmental Impact
- Hardware Type: A100 GPU
- Hours used: X
- Cloud Provider: [More Information Needed]
- Compute Region: [More Information Needed]
- Carbon Emitted: [More Information Needed]
Technical Specifications [optional]
Model Architecture and Objective
- The model is based on T5-small architecture fine-tuned for machine translation.
Compute Infrastructure
Hardware
- Training was performed on a single A100 GPU
Software
- Transformers, Datasets, PEFT, Accelerate, Evaluate, Torch
Citation [optional]
BibTeX:
@misc{hinglish_translation,
author = {Your Name},
title = {English to Hinglish Translation Model},
year = {2025},
url = {https://huggingface.co/rairashmi/hinglish_translation_lora}
}
Glossary [optional]
- Hinglish: A mix of Hindi and English written in Latin script.
More Information [optional]
For further details, check out the Hugging Face Model Page.
Model Card Authors [optional]
- [Your Name or Organization]
Model Card Contact
For any issues or questions, contact [Your Contact Information].