metadata
language: en
license: apache-2.0
tags:
- text-classification
- intent-detection
- distilbert
- nlu
datasets:
- custom
metrics:
- accuracy
- f1
pipeline_tag: text-classification
DistilBERT NLU Intent Classification
Fine-tuned DistilBERT model for intent classification in Natural Language Understanding (NLU) systems.
Model Details
- Base Model: distilbert-base-uncased
- Task: Intent Classification (Sequence Classification)
- Number of Labels: 8
- Framework: PyTorch + Transformers
Supported Intents
| ID | Intent | Description |
|---|---|---|
| 0 | BILLING_ISSUE | Problems with bills or charges |
| 1 | CANCEL_SUBSCRIPTION | Cancel service requests |
| 2 | CHECK_BALANCE | Balance inquiry |
| 3 | GOODBYE | Farewell messages |
| 4 | GREETING | Hello/welcome messages |
| 5 | MODIFY_PROFILE | Update account details |
| 6 | ROAMING_ACTIVATION | Enable roaming |
| 7 | ROAMING_DEACTIVATION | Disable roaming |
Usage
from transformers import pipeline
classifier = pipeline("text-classification", model="sidde/distilbert-nlu-intent-classification")
# Single prediction
result = classifier("I want to check my balance")
print(result)
# [{"label": "CHECK_BALANCE", "score": 0.98}]
Training Details
- Dataset: 772 examples (custom intent dataset)
- Train/Eval Split: 80/20 with stratification
- Epochs: 10
- Batch Size: 16
- Learning Rate: 2e-5
- Hardware: NVIDIA L4 GPU on OpenShift AI
Deployment
This model is deployed on OpenShift AI using KServe.
License
Apache 2.0