---
license: mit
datasets:
- cngchis/Support-Ticket-Router-12K-Cleaned
language:
- en
metrics:
- f1
- accuracy
- precision
- recall
base_model:
- unsloth/Phi-4-mini-instruct
new_version: cngchis/phi4-mini-intent
pipeline_tag: text-classification
library_name: transformers
---

# Intent Classification Model

## Model Description

This repository contains a **fine-tuned Transformer model** for **intent classification**.

The model is built using Hugging Face `transformers` and stored in **safetensors format**, enabling efficient and safe loading.

It predicts an intent label from input text for tasks such as chatbot understanding, ticket routing, and text categorization.

---

## How to Use

### Install dependencies

```bash
pip install transformers==4.57.6
```

---

### Load model
```bash
from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

model_path = "cngchis/phi4-mini-intent"

tokenizer = AutoTokenizer.from_pretrained(model_path)
model = AutoModelForSequenceClassification.from_pretrained(model_path)

text = "I cannot log into my account"

inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True)

with torch.no_grad():
    outputs = model(**inputs)

logits = outputs.logits
predicted_class = torch.argmax(logits, dim=1).item()

print(predicted_class)
```

---

### Input Format (Recommended)

```json
"I want to reset my password"
```

---

### Output Format

The model outputs a class index, which can be mapped to intent labels:

```json
3  → password_reset
1  → login_issue
5  → payment_problem
```

(You should define label mapping in your application.)

---

### Model Details

- Architecture: Transformer-based classification model
- Task: Intent classification
- Format: PyTorch (safetensors)
- Library: Hugging Face Transformers
- Input: Natural language text
- Output: Single intent class

---

### Notes
- Best performance when input format matches training data
- Requires label mapping for interpretation
- Works with GPU
- Supports batch inference via Transformers

---

### Limitations

Not suitable for generative tasks
Sensitive to domain shift (out-of-distribution text)
Requires consistent intent label schema

---

### Acknowledgements

Built using:

Hugging Face Transformers
PyTorch
Safetensors format