Add Support for Larger Models (Llama 2, Mistral, Mixtral,etc)

[Shehrozkashif]

📌 Background / Context

Extending support to large-scale models requires careful memory management and quantization. This allows the system to handle modern LLMs efficiently.

🎯 Objectives

• Support larger LLMs with optimized memory usage
• Enable low-bit training and inference
• Maintain flexibility for different models and fine-tuning approaches

✅ Tasks

• Add 4-bit/8-bit quantization support using bitsandbytes
• Implement gradient accumulation for large batch sizes
• Add model-specific prompt templates
• Enable LoRA/QLoRA fine-tuning options
• Update documentation with setup instructions
• Test larger models for inference and training