mjpsm/Llama-3.2-1B-Instruction-Chatbot-LoRA

Llama 3.2 1B Instruction Chatbot (LoRA Adapter)

Overview

This repository contains a LoRA adapter fine-tuned from Meta's Llama 3.2 1B Instruct model using supervised instruction tuning.

This repository does not contain the base model. Users must have access to the original base model before loading this adapter.

The purpose of this project was educational—to learn the complete workflow of instruction tuning, QLoRA, PEFT, model evaluation, and deployment using Hugging Face.

---

Base Model

• meta-llama/Llama-3.2-1B-Instruct

---

Training Dataset

Approximately 40,000 instruction-response examples.

Dataset format:

• Instruction
• Optional Input
• Response

The dataset was converted into a conversational chat format using the Llama chat template prior to training.

---

Training Configuration

Fine-tuning Method

• QLoRA
• PEFT
• 4-bit NF4 Quantization

Frameworks

• Transformers
• TRL
• PEFT
• BitsAndBytes
• Accelerate

---

Hyperparameters

Parameter	Value
Epochs	1
Batch Size	2
Gradient Accumulation	4
Learning Rate	2e-4
Optimizer	paged_adamw_8bit
Scheduler	Cosine
Precision	FP16

---

Training Results

Metric	Value
Training Loss	0.15088
Runtime	6789.63 seconds
Global Steps	5000
Samples / Second	5.89
Steps / Second	0.736

---

Intended Uses

• Instruction following
• Conversational AI
• Educational projects
• Fine-tuning demonstrations
• Research

---

Limitations

This adapter was created for educational purposes.

It has not undergone extensive benchmark evaluation or safety alignment beyond the underlying base model.

Responses may contain inaccuracies or hallucinations.

---

Loading the Adapter

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base_model = AutoModelForCausalLM.from_pretrained(
    "meta-llama/Llama-3.2-1B-Instruct"
)

model = PeftModel.from_pretrained(
    base_model,
    "mjpsm/Llama-3.2-1B-Instruction-Chatbot-LoRA"
)

tokenizer = AutoTokenizer.from_pretrained(
    "meta-llama/Llama-3.2-1B-Instruct"
)

---

Citation

If you use this adapter in your work, please cite this repository.