---
base_model: Qwen/Qwen2.5-1.5B-Instruct
library_name: peft
pipeline_tag: text-generation
license: mit
tags:
- base_model:adapter:Qwen/Qwen2.5-1.5B-Instruct
- lora
- peft
- transformers
- text-generation
- personalization
---

# LoRA Project

This repository contains a PEFT LoRA adapter for personalized email-style generation.
The adapter was trained from `Qwen/Qwen2.5-1.5B-Instruct` using project-specific
style examples exported from the ACM personalization pipeline.

This is not a standalone full model. Load it together with the base model.

## Base Model

`Qwen/Qwen2.5-1.5B-Instruct`

## Intended Use

The adapter is intended for class project experimentation with personalized writing
style generation. It can be used with the companion GitHub project to generate
email-style responses from a selected user's profile and prompt.

Project repo:
`https://github.com/AryanAGit/LLM_Personalization---ACM-AI-Team4`

## Loading The Adapter

```python
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base_model = "Qwen/Qwen2.5-1.5B-Instruct"
adapter_id = "alchin2/lora-project"

tokenizer = AutoTokenizer.from_pretrained(base_model)
model = AutoModelForCausalLM.from_pretrained(base_model)
model = PeftModel.from_pretrained(model, adapter_id)
```

## Local App Usage

From the GitHub project directory, run:

```bash
python3 project.py generate \
  --history data/processed/user_email_history.json \
  --backend peft \
  --base-model Qwen/Qwen2.5-1.5B-Instruct \
  --adapter-path alchin2/lora-project \
  --prompt "Write an email asking Wendy to send the files by Wednesday."
```

Or start the local web app:

```bash
python3 run_app.py
```

Then open `http://127.0.0.1:8787`.

## Evaluation

Recommended evaluation methods for this project:

- Automatic metrics from the project evaluation command, including style distance,
  greeting/signoff matching, length ratio, word overlap, and profile copy risk.
- LaMP-style personalization evaluation, comparing outputs with and without user
  profile context.
- Blind LLM or human ranking across base model, RAG, LoRA, and RAG+LoRA outputs.

## Limitations

The adapter is trained for style imitation experiments and may overfit small user
profiles. Outputs should be checked for prompt faithfulness, fluency, and copying
from training examples.