gptmed 0.3.2

A lightweight GPT-based language model framework for training custom question-answering models on any domain

GptMed 🤖

A lightweight GPT-based language model framework for training custom question-answering models on any domain. This package provides a transformer-based GPT architecture that you can train on your own Q&A datasets - whether it's casual conversations, technical support, education, or any other domain.

📖 Complete User Manual | Quick Start

New to GptMed? Check out the step-by-step User Manual for a complete guide on training your own model!

Features

• 🧠 Custom GPT Architecture: Lightweight transformer model for any Q&A domain
• 🎯 Domain-Agnostic: Train on any question-answering dataset (casual chat, tech support, education, etc.)
• ⚡ Fast Inference: Optimized for quick question answering
• 🔧 Flexible Training: Easy to train on your own custom datasets
• 📦 Lightweight: Small model size suitable for edge deployment
• 🛠️ Complete Toolkit: Includes tokenizer training, model training, and inference utilities

Table of Contents

• Features
• Installation
• Quick Start
• Package Structure
  • Core Modules
  • Model Components
  • Training Components
  • Inference Components
  • Data Processing
  • Utilities
• Model Architecture
• Configuration
• Documentation
• Performance
• Examples
• Contributing
• License
• Support

Installation

From PyPI (Recommended)

pip install gptmed

From Source

git clone https://github.com/sigdelsanjog/gptmed.git
cd gptmed
pip install -e .

With Optional Dependencies

# For development
pip install gptmed[dev]

# For training
pip install gptmed[training]

# All dependencies
pip install gptmed[dev,training]

Quick Start

Inference (Generate Answers)

from gptmed.inference.generator import TextGenerator
from gptmed.model.architecture import GPTTransformer
from gptmed.model.configs.model_config import get_small_config

# Load model
config = get_small_config()
model = GPTTransformer(config)

# Load your trained checkpoint
# model.load_state_dict(torch.load('path/to/checkpoint.pt'))

# Create generator
generator = TextGenerator(
    model=model,
    tokenizer_path='path/to/tokenizer.model'
)

# Generate answer
question = "What's your favorite programming language?"
answer = generator.generate(
    prompt=question,
    max_length=100,
    temperature=0.7
)

print(f"Q: {question}")
print(f"A: {answer}")

Using Command Line

# Generate answers
gptmed-generate --prompt "How do I train a custom model?" --max-length 100

# Train model
gptmed-train --model-size small --num-epochs 10 --batch-size 16

Training Your Own Model

from gptmed.training.train import main
from gptmed.configs.train_config import get_default_config
from gptmed.model.configs.model_config import get_small_config

# Configure training
train_config = get_default_config()
train_config.batch_size = 16
train_config.num_epochs = 10
train_config.learning_rate = 3e-4

# Start training
main()

Model Architecture

The model uses a custom GPT-based transformer architecture:

• Embedding: Token + positional embeddings
• Transformer Blocks: Multi-head self-attention + feed-forward networks
• Parameters: ~10M (small), ~50M (medium)
• Context Length: 512 tokens
• Vocabulary: Custom SentencePiece tokenizer trained on your data

Configuration

Model Sizes

from gptmed.model.configs.model_config import (
    get_tiny_config,   # ~2M parameters - for testing
    get_small_config,  # ~10M parameters - recommended
    get_medium_config  # ~50M parameters - higher quality
)

Training Configuration

from gptmed.configs.train_config import TrainingConfig

config = TrainingConfig(
    batch_size=16,
    learning_rate=3e-4,
    num_epochs=10,
    warmup_steps=100,
    grad_clip=1.0
)

Package Structure

Core Modules

The gptmed package contains the following main modules:

gptmed/
├── model/                  # Model architecture and configurations
├── inference/              # Text generation and sampling
├── training/               # Training loops and datasets
├── tokenizer/              # Tokenizer training and data processing
├── data/                   # Data parsers and formatters
├── configs/                # Training configurations
└── utils/                  # Utilities (checkpoints, logging)

Model Components

gptmed.model.architecture - GPT Transformer Implementation

• GPTTransformer - Main model class
• TransformerBlock - Individual transformer layers
• MultiHeadAttention - Attention mechanism
• FeedForward - Feed-forward networks
• RoPEPositionalEncoding - Rotary position embeddings

gptmed.model.configs - Model Configurations

• get_tiny_config() - ~2M parameters (testing)
• get_small_config() - ~10M parameters (recommended)
• get_medium_config() - ~50M parameters (high quality)
• ModelConfig - Custom configuration class

Training Components

gptmed.training - Training Pipeline

• train.py - Main training script (CLI: gptmed-train)
• Trainer - Training loop with checkpointing
• TokenizedDataset - PyTorch dataset for tokenized data
• create_dataloaders() - DataLoader creation utilities

gptmed.configs - Training Configurations

• TrainingConfig - Training hyperparameters
• get_default_config() - Default training settings
• get_quick_test_config() - Fast testing configuration

Inference Components

gptmed.inference - Text Generation

• TextGenerator - Main generation class
• generator.py - CLI command (CLI: gptmed-generate)
• sampling.py - Sampling strategies (top-k, top-p, temperature)
• decoding_utils.py - Decoding utilities
• GenerationConfig - Generation parameters

Data Processing

gptmed.tokenizer - Tokenizer Training & Data Processing

• train_tokenizer.py - Train SentencePiece tokenizer
• tokenize_data.py - Convert text to token sequences
• SentencePiece BPE tokenizer support

gptmed.data.parsers - Data Parsing & Formatting

• MedQuADParser - XML Q&A parser (example)
• CausalTextFormatter - Format Q&A pairs for training
• FormatConfig - Formatting configuration

Utilities

gptmed.utils - Helper Functions

• checkpoints.py - Model checkpoint management
• logging.py - Training metrics logging

---

Detailed Project Structure

gptmed/
├── model/
│   ├── architecture/
│   │   ├── gpt.py              # GPT transformer model
│   │   ├── attention.py        # Multi-head attention
│   │   ├── feedforward.py      # Feed-forward networks
│   │   └── embeddings.py       # Token + positional embeddings
│   └── configs/
│       └── model_config.py     # Model size configurations
├── inference/
│   ├── generator.py            # Text generation (CLI command)
│   ├── sampling.py             # Sampling strategies
│   ├── decoding_utils.py       # Decoding utilities
│   └── generation_config.py    # Generation parameters
├── training/
│   ├── train.py                # Main training script (CLI command)
│   ├── trainer.py              # Training loop
│   ├── dataset.py              # PyTorch dataset
│   └── utils.py                # Training utilities
├── tokenizer/
│   ├── train_tokenizer.py      # Train SentencePiece tokenizer
│   └── tokenize_data.py        # Tokenize text data
├── data/
│   └── parsers/
│       ├── medquad_parser.py   # Example XML parser
│       └── text_formatter.py   # Q&A text formatter
├── configs/
│   └── train_config.py         # Training configurations
└── utils/
    ├── checkpoints.py          # Model checkpointing
    └── logging.py              # Training logging

Command-Line Interface

The package provides two main CLI commands:

# Train a model
gptmed-train --model-size small --num-epochs 10 --batch-size 16

# Generate text
gptmed-generate --prompt "Your question?" --max-length 100

Requirements

• Python >= 3.8
• PyTorch >= 2.0.0
• sentencepiece >= 0.1.99
• numpy >= 1.24.0
• tqdm >= 4.65.0

Documentation

📚 Complete User Manual - Step-by-step guide for training your own model

Quick Links

• User Manual - Start here! Complete training pipeline guide
• Architecture Guide - Understanding the model architecture
• Deployment Guide - Publishing to PyPI
• Changelog - Version history

Performance

Model Size	Parameters	Training Time	Inference Speed
Tiny	~2M	2 hours	~100 tokens/sec
Small	~10M	8 hours	~80 tokens/sec
Medium	~50M	24 hours	~50 tokens/sec

Tested on GTX 1080 8GB

Examples

Medical Question Answering

# Example 1: Symptoms inquiry
question = "What are the early signs of Alzheimer's disease?"
answer = generator.generate(question, temperature=0.7)

# Example 2: Treatment information
question = "How is Type 2 diabetes treated?"
answer = generator.generate(question, temperature=0.6)

# Example 3: Medical definitions
question = "What is hypertension?"
answer = generator.generate(question, temperature=0.5)

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

1. Fork the repository
2. Create your feature branch (git checkout -b feature/AmazingFeature)
3. Commit your changes (git commit -m 'Add some AmazingFeature')
4. Push to the branch (git push origin feature/AmazingFeature)
5. Open a Pull Request

Citation

If you use this model in your research, please cite:

@software{llm_med_2026,
  author = {Sanjog Sigdel},
  title = {GptMed: A custom causal question answering general purpose GPT Transformer Architecture Model},
  year = {2026},
  url = {https://github.com/sigdelsanjog/gptmed}
}

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• MedQuAD dataset creators
• PyTorch team

Disclaimer

⚠️ Medical Disclaimer: This model is for research and educational purposes only. It should NOT be used for actual medical diagnosis or treatment decisions. Always consult qualified healthcare professionals for medical advice.

Support

• � User Manual - Complete step-by-step training guide
• �📫 Issues: GitHub Issues
• 💬 Discussions: GitHub Discussions
• 📧 Email: [e-mail]

Changelog

See CHANGELOG.md for version history.

---

Made with ❤️ for learning purpose