This repository contains a LoRA (Low-Rank Adaptation) adapter fine-tuned on top of the base model Qwen3-4B-Instruct-2507 for user intent classification.

The model is designed to analyze the latest user utterance within a conversation and classify it into a predefined set of intention categories. These categories are customizable and defined by business users (e.g., sales or customer service managers).

The adapter is lightweight and intended to be merged with the base model at inference time.

Developed by: Li Tuo
Model type: Causal Language Model (LLM) with LoRA adapter for classification
Language(s) (NLP): Chinese (primary), English (partial support)
License: Apache-2.0
Finetuned from model: Qwen3-4B-Instruct-2507

Uses

This model is intended to be used as a self-hosted intent classification module within AI-powered customer engagement systems, such as:

AI marketing agents
Customer service chatbots
Phone call automation systems

It is a core component of the project: 👉 AI CS agent generator

Direct Use

Given a prompt that includes customized intention classes and a multi-turn conversation, the model:

Focuses on the latest user input
Extracts a structured intent label
Outputs a predefined intent ID

Bias, Risks, and Limitations

Performance depends on quality of synthetic data
May degrade on unseen domains (e.g. conversation with non commercial event topics)

Recommendations

The fine-tuned model is intened to serve the AI customer service agent focused on commericial event engagement calls.

How to Get Started with the Model

Download the project repository, and set up the environment (Python 3.11 recommended):

Bash

git clone https://github.com/lituokobe/Qwen3-Fine-Tuning.git

cd Qwen3-Fine-Tuning

pip install -r requirements

Download the base model and adapter files to their respective locations:

Bash

huggingface-cli download Qwen/Qwen3-4B-Instruct-2507 \
  --local-dir models/Qwen/Qwen3-4B-Instruct-2507

huggingface-cli download lituokobe/Qwen3-4B-Instruct-2507-LoRA-Intent-Classifier \
  --local-dir Qwen3_reg/lora_adapter_lr1e-4_1epoch

Deploy in the same environment:

shell

vllm serve models/Qwen/Qwen3-4B-Instruct-2507 \
    --enable-lora \
    --lora-modules qwen-finetune= Qwen3_reg/lora_adapter_lr1e-4_1epoch \
    --max-lora-rank 64 \
    --max-model-len 4096 \
    --trust-remote-code \
    --dtype float16 \
    --port 8000 \
    --served-model-name qwen3-4b

After the deployment is successful, you can test the model using a simple curl request:

shell

curl http://localhost:8000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen3-4b",
    "messages": [
      {
        "role": "user",
        "content": "Please take any sample from the training data or adapt it with the correct format"
      }
    ],
    "temperature": 0.0,
    "max_tokens": 50
  }'

The model is expected to return a structured dictionary (JSON) with the following schema:

JSON

{
  "input_summary": "A concise summary of the user's latest message.",
  "intention_id": "short_intent_label"
}

Training Details

Training Data

LoRA-Samples-Intention-Classifier

Training Procedure

Preprocessing [optional]

[More Information Needed]

Training Hyperparameters

Training regime: [More Information Needed]

Speeds, Sizes, Times [optional]

[More Information Needed]

Evaluation

Testing Data, Factors & Metrics

Testing Data

[More Information Needed]

Factors

[More Information Needed]

Metrics

[More Information Needed]

Results

[More Information Needed]

Summary

Model Examination [optional]

[More Information Needed]

Environmental Impact

Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019).

Hardware Type: [More Information Needed]
Hours used: [More Information Needed]
Cloud Provider: [More Information Needed]
Compute Region: [More Information Needed]
Carbon Emitted: [More Information Needed]

Technical Specifications [optional]

Model Architecture and Objective

[More Information Needed]

Compute Infrastructure

[More Information Needed]

Hardware

[More Information Needed]

Software

[More Information Needed]

Citation [optional]

BibTeX:

[More Information Needed]

APA:

[More Information Needed]

Glossary [optional]

[More Information Needed]

More Information [optional]

[More Information Needed]

Model Card Authors [optional]

[More Information Needed]

Model Card Contact

[More Information Needed]

Framework versions

PEFT 0.18.1