Sheikh Bengali AI Model

Model Description

Sheikh is a Bengali (Bangla) language AI model trained for instruction following and conversational tasks. Built on top of Microsoft's DialoGPT-medium, this model has been fine-tuned with Bengali instruction-following data to understand and generate responses in Bengali language.

Model Details

• Model Type: Language Model, Text Generation
• Architecture: GPT-2 based (DialoGPT-medium)
• Base Model: microsoft/DialoGPT-medium
• Parameters: 355M
• Language: Bengali (Bangla)
• Training Data: Alpaca Bangla instruction dataset
• Model Size: 1.4GB
• License: Apache 2.0

Intended Use

This model is designed for:

• Bengali language text generation
• Instruction following and question answering
• Educational content creation
• Cultural and historical knowledge responses
• General conversational AI in Bengali

Usage

from transformers import AutoTokenizer, AutoModelForCausalLM

# Load the model
tokenizer = AutoTokenizer.from_pretrained("megharudushi/Sheikh")
model = AutoModelForCausalLM.from_pretrained("megharudushi/Sheikh")

# Generate Bengali response
input_text = "বাংলাদেশের রাজধানী কী?"
inputs = tokenizer.encode(input_text, return_tensors="pt")
outputs = model.generate(inputs, max_length=150, temperature=0.8)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response)

Example Prompts

• Educational: "গণিতের মৌলিক নীতি বলুন"
• Cultural: "বাংলা সাহিত্যের বিখ্যাত কবি কারা?"
• General: "স্বাস্থ্যকর থাকার উপায় বলুন"
• Historical: "বাংলাদেশের স্বাধীনতার ইতিহাস বর্ণনা করুন"

Model Performance

• Supports Bengali language understanding and generation
• Trained on Bengali instruction-following dataset
• Optimized for educational and conversational contexts
• Cultural knowledge preservation for Bengali language

Limitations

• Trained primarily on Bengali instruction data
• May have limitations in very specialized domains
• Performance depends on input quality and clarity
• Model size limited by computational resources

Training Details

• Base Model: microsoft/DialoGPT-medium
• Fine-tuning Data: Alpaca Bangla dataset
• Training Approach: Instruction following
• Language Focus: Bengali (Bangla) language

Citation

If you use this model, please cite:

@misc{SheikhBengaliAI,
  title={Sheikh Bengali AI Model},
  author={megharudushi},
  year={2025},
  url={https://huggingface.co/megharudushi/Sheikh},
  note={Bengali language instruction-following model based on DialoGPT-medium}
}

License

This model is released under the Apache 2.0 License.

Contributing

This model is part of the Bengali AI initiative to make Bengali language AI more accessible to the community.

---

Created: December 21, 2025
Repository: https://huggingface.co/megharudushi/Sheikh
Base Model: microsoft/DialoGPT-medium
Language: Bengali (Bangla)