llama-3.2-3b-qlora-myproject

🦙 Llama 3.2 3B — QLoRA Fine-Tuned

Parameter-efficient instruction fine-tuning of unsloth/Llama-3.2-3B-Instruct using QLoRA.

✨ Highlights

• 4-bit quantized base model
• LoRA-based parameter-efficient fine-tuning
• NF4 quantization
• Unsloth optimized training
• Hugging Face TRL SFTTrainer
• PEFT adapters
• Designed for low-VRAM experimentation

---

🧠 Model Overview

This model is a QLoRA fine-tuned adapter based on:

unsloth/Llama-3.2-3B-Instruct

The base model remains frozen while lightweight LoRA adapters are trained on the instruction dataset.

Training Pipeline

Pretrained Llama 3.2 3B
        │
        ▼
   4-bit Quantization
        │
        ▼
   Frozen Base Model
        │
        ▼
      LoRA
        │
        ▼
  Supervised Fine-Tuning
        │
        ▼
  Fine-Tuned Adapter