lokisol/Qwen2.5model

Lokesh B Personal Profile Assistant

A small personalized profile assistant fine-tuned from Qwen/Qwen2.5-0.5B-Instruct using LoRA and 4-bit quantization.

Profile

• Name: Lokesh B
• College: Dayananda Sagar University
• Program: MSc Data Science

Training

The original notebook uses:

• Base model: Qwen/Qwen2.5-0.5B-Instruct
• LoRA rank (r): 16
• LoRA alpha: 32
• LoRA dropout: 0.05
• Target modules: q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj
• 4-bit NF4 quantization
• Double quantization enabled
• Maximum sequence length: 512
• Epochs: 10
• Batch size: 2
• Gradient accumulation: 4
• Learning rate: 2e-4

Dataset

The repository contains the same 12 profile-oriented training examples from the original notebook in data/training_data.json.

Files

• train.py — LoRA fine-tuning script
• inference.py — command-line inference
• app.py — optional Gradio interface
• data/training_data.json — training examples
• requirements.txt — Python dependencies

Training

pip install -r requirements.txt
python train.py

The LoRA adapter is saved in:

./lokesh_finetuned_model

Inference

After training:

python inference.py

Example questions:

Who is Lokesh B?
Where does Lokesh B study?
What is Lokesh B studying?
Give me Lokesh B details.

Gradio interface

Install Gradio:

pip install gradio

Then run:

python app.py

Important note

This repository contains a LoRA adapter rather than a complete copy of the Qwen base model. The base model is downloaded from Hugging Face when the scripts run.

The assistant should be treated as a student/profile demonstration, not as a source of information beyond the profile data used for training.