---
license: apache-2.0
datasets:
- dair-ai/emotion
metrics:
- accuracy
base_model:
- microsoft/deberta-v3-base
pipeline_tag: text-classification
library_name: transformers
---
# DeBERTa v3 Emotion Classifier


## Model description
This repository contains a DeBERTa v3 base model fine-tuned for emotion classification on the `dair-ai/emotion` dataset. The model is intended for short-text emotion labeling and was finetuned with standard Trainer-based training on Google Colab / Drive.

## Use cases
- Classifying short texts into emotion categories for downstream workflows (analytics, moderation, UX signals).
- Human-in-the-loop pipelines where low-confidence outputs trigger clarification.

## Dataset
- `dair-ai/emotion` (public dataset for emotion labeling).

## Training summary
- Base model: `microsoft/deberta-v3-base`
- Fine-tuning method: full fine-tuning (Trainer)
- Number of labels: 6
- Training environment: Google Colab 

## intended use
This model is intended to support emotion classification tasks. It may produce incorrect or biased outputs when used on out-of-distribution text, long-form inputs, or languages it was not trained on. Use a confidence-based fallback for important decisions and include human review for high-stakes applications.

## How to load
```python
from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch
repo_id = ragunath-ravi/deberta-v3-emotion-classifier
tokenizer = AutoTokenizer.from_pretrained(repo_id)
model = AutoModelForSequenceClassification.from_pretrained(repo_id)
device = "cuda" if torch.cuda.is_available() else "cpu"
model.to(device)
model.eval()
```

## Example inference
```python
from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch
import torch.nn.functional as F
tokenizer = AutoTokenizer.from_pretrained(ragunath-ravi/deberta-v3-emotion-classifier)
model = AutoModelForSequenceClassification.from_pretrained(ragunath-ravi/deberta-v3-emotion-classifier)
inputs = tokenizer("I am so happy today!", return_tensors="pt", truncation=True, padding=True)
with torch.no_grad():
    logits = model(**inputs).logits
probs = F.softmax(logits, dim=-1).cpu().numpy()[0]
print(probs)
```

## Acknowledgements
- This model is based on Microsoft DeBERTa v3 base.
- Dataset: `dair-ai/emotion`.
- Transformers library: Hugging Face `transformers`.

## Model demo : [demo](https://huggingface.co/spaces/ragunath-ravi/deberta-emotion-classifier)