YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
Marwari AI Language Model
Overview
Marwari AI is a multilingual language model focused on understanding and generating Pakistani Marwari dialect along with related languages.
The model was fine-tuned using a custom multilingual dataset containing:
- Marwari
- Romanized Marwari
- Urdu
- Hindi
- English
- Sindhi (partial)
The goal of this project is to enable AI support for Marwari, which is currently a low-resource language with very limited NLP tools.
Model Details
Attribute| Value Model Name| marwari_ai Author| Abdul Raheem Framework| Hugging Face Transformers Model Type| Causal Language Model Base Architecture| Transformer (GPT-style) Weights Format| safetensors Export Format| LiteRT + HuggingFace License| Apache-2.0
Files Included
The repository contains:
config.json model.safetensors tokenizer.json tokenizer_config.json generation_config.json training_args.bin marwari_model.litertlm
These files allow the model to run with Transformers, Torch, or LiteRT environments.
Training Data
The training dataset contains multilingual aligned text samples such as:
Example:
Urdu
میں کل بازار گیا تھا
Marwari
ھون کال بازار ڳيو ھتو
Romanized Marwari
hon kaal bazar giyo hto
English
I went to the market yesterday
The dataset includes:
- conversational sentences
- translation style examples
- short phrases
- manually corrected language pairs
This approach helps the model understand language relationships between Urdu, Hindi, and Marwari.
Intended Use
This model can be used for:
• Marwari text generation • multilingual translation experiments • conversational AI • language preservation research • linguistic experiments on low-resource languages
Possible applications include:
- chatbots
- translation tools
- educational tools
- voice assistants
Example Usage
from transformers import AutoTokenizer, AutoModelForCausalLM
model_name = "abdulraheemn/marwari_ai"
tokenizer = AutoTokenizer.from_pretrained(model_name) model = AutoModelForCausalLM.from_pretrained(model_name)
prompt = "میں کل بازار گیا تھا"
inputs = tokenizer(prompt, return_tensors="pt")
output = model.generate( **inputs, max_length=80, temperature=0.8, do_sample=True )
print(tokenizer.decode(output[0], skip_special_tokens=True))
Limitations
Marwari is a low-resource language, therefore:
- dataset size is still limited
- spelling variations exist in Roman Marwari
- dialect differences may affect results
Future improvements will increase the dataset size and language coverage.
Future Improvements
Planned updates for this project:
• larger Marwari dataset • speech training using Marwari audio • better Romanized Marwari support • mobile AI integration • real-time translation tools
Ethical Considerations
This model was trained only on publicly available resources and community provided data.
Users should verify model outputs before using them in critical systems.
License
This model is released under the Apache License 2.0.
You are free to:
- use
- modify
- distribute
- build commercial applications
as long as you follow the Apache 2.0 license terms.
License: https://www.apache.org/licenses/LICENSE-2.0
Author
Abdul Raheem
Independent developer working on:
- low resource language AI
- multilingual NLP
- community language preservation
- Downloads last month
- -