Add trainable LLM AI infrastructure with Qwen2.5 1B models

[Tawns-lab]

Implement comprehensive training setup using Unsloth framework for efficient fine-tuning of 1 billion parameter models. This adds multi-capability and adaptability features for LLM training on consumer hardware.

Features:

• Complete training pipeline for Qwen2.5-0.5B/1.5B models
• 2x faster training with 70% less VRAM via Unsloth
• QLoRA, LoRA, and full fine-tuning support
• Single GPU training (6-8GB VRAM minimum)
• Interactive inference and batch processing tools
• Configurable presets for different use cases

Files added:

• llm-training/train_qwen_1b.py: Main training script with comprehensive config
• llm-training/quickstart.py: 5-minute quick start for testing
• llm-training/inference.py: Interactive and batch inference tool
• llm-training/config.py: Training configuration templates and presets
• llm-training/setup.sh: Automated environment setup script
• llm-training/requirements.txt: Python dependencies
• llm-training/README.md: Complete documentation and guides

Supports:

• Multi-modal capabilities (text, code, reasoning)
• Custom datasets and fine-tuning workflows
• Export to GGUF, GPTQ formats
• Experiment tracking (WandB, TensorBoard)

🤖 Generated with Claude Code