qwen2-1.5b-customer-support-sft-lora-v1

Qwen2-1.5B Customer Support SFT LoRA (v1)

This repository contains a LoRA adapter fine-tuned on Qwen2-1.5B-Instruct for structured customer support ticket classification.

⚠️ This repository contains only the LoRA adapter weights, not the base model.

The base model used: Qwen/Qwen2-1.5B-Instruct (Apache-2.0 licensed)

Model Overview

• Base Model: Qwen2-1.5B-Instruct
• Fine-tuning Method: Supervised Fine-Tuning (SFT)
• Adapter Method: LoRA (PEFT)
• Task: Customer Support Ticket Classification
• Output Format: Text-based category prediction

---

Intended Use

This model is intended for:

• Customer support workflow prototyping
• Structured response generation
• Research and educational purposes

Not intended for:

• Production-critical systems without further validation
• Legal, financial or medical advice

---

Training Details

• Base Model: Qwen2-1.5B-Instruct
• Fine-tuning Method: Supervised Fine-Tuning (SFT)
• Adapter Type: LoRA
• LoRA Rank: 16
• Epochs: 5
• Learning Rate: 1e-4
• Framework: Transformers + TRL + PEFT

Dataset

Trained on a synthetic customer support dataset generated for research and educational purposes.

No proprietary or real customer data was used.

---

How to Use

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base_model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen2-1.5B-Instruct")
model = PeftModel.from_pretrained(base_model, "mukizone/qwen2-1.5b-customer-support-sft-lora-v1")

tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen2-1.5B-Instruct")

prompt = """
### Instruction:
You are an customer support classifier.
Return ONLY a valid JSON object with:
- category
- priority
- response

Allowed categories:
- delivery_issue
- refund_request
- damaged_item
- product_query
You MUST choose only from these.

### Input:
My package arrived damaged and I want a replacement.

### Output:

"""

inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Limitations

• May hallucinate unsupported categories
• Sensitive to prompt phrasing
• Not evaluated for safety-critical environments

License

This adapter is released under Apache-2.0. The base model license applies to the underlying weights.

Users must comply with the base model’s license: https://huggingface.co/Qwen/Qwen2-1.5B-Instruct