LoRA Streamlit UI - Professional AI Training & Inference

A modular, high-performance toolkit for LoRA training (via Google Colab/Cloud) and optimized Intel-native inference (via OpenVINO).

📂 Project Structure

• src/app.py: Main Streamlit UI entry point.
• src/engine/: Core inference engines (Standard & Intel-optimized).
• src/utils/: Shared helper functions and memory management.
• notebooks/: Jupyter notebooks for quickstart and custom training.
• scripts/: Setup, installation, and utility scripts.
• docs/: Guides and walkthroughs.
• tests/: Unit and integration testing suite.
• data/: Local storage for training datasets.
• outputs/: Default directory for generated images and logs.

🚀 Quick Start (Local)

1. Clone & Setup:

   git clone https://github.com/melbinjp/Lora_trainer
   cd Lora_trainer
   # Windows
   .\scripts\setup.bat
   # Linux/macOS
   bash scripts/setup.sh

2. Run the App:

   streamlit run src/app.py

☁️ Google Colab (Cloud Training)

1. Open colab_quickstart.ipynb in Google Colab.
2. Run the cells to launch the UI and begin training.
3. The UI will be accessible via a secure trycloudflare.com tunnel.

🖥️ Intel-Native Optimized Inference

This project includes specialized support for Intel Core Ultra systems (Meteor Lake/Lunar Lake) using OpenVINO.

• Stable Hybrid Mode: Automatically partitions the Flux.1 model between CPU and iGPU to ensure stability on 16GB RAM systems.
• Real-Time UI Progress: Streamlit UI now displays live progress bars and status updates during inference.
• Dynamic Prompts: Generate images directly from the UI or via CLI:

  python src/engine/intel_inference.py STABLE_HYBRID --prompt "your prompt" --output outputs/result.png

🌐 API & Distributed Processing

A lightweight API is included to allow programmatic access to the inference engine. This is ideal for distributed networks, batch processing, or integrating with other applications.

1. Start the API Server:

   python src/api.py

2. Send a Generation Request (example using curl or Postman):

   curl -X POST "http://localhost:8000/generate" -H "Content-Type: application/json" -d '{"prompt": "A futuristic city", "model": "OpenVINO/FLUX.1-schnell-int4-ov", "hardware_target": "STABLE_HYBRID"}' --output result.png

🛠️ Features & Tools

Dataset Engineering

• Auto-Captioning: Uses BLIP to automatically generate captions for your training images.
• Export: Packages your dataset into an optimized zip file ready for cloud training.

Memory Management

• Aggressive RAM Flush: Includes utilities to clear system memory before intensive inference runs.
• Resource Detection: Automatically detects available RAM and VRAM to suggest optimal settings.

Cloud Integration

• Hugging Face: Seamlessly download/upload models using HF Hub integration.
• Cloud Storage: Mount Google Drive, S3, or OneDrive for persistent model storage.

🧪 Testing

Run the test suite to ensure everything is configured correctly:

pytest tests/unit/test_logic.py

📜 Documentation

For more detailed information, see:

• Walkthrough Guide
• Setup Instructions

---

Created with ❤️ for AI researchers and enthusiasts.