distilbert-goemotions

🧠 DistilBERT Emotion Classifier (GoEmotions, Multi-Label)

This model is a fine-tuned version of distilbert-base-uncased on the GoEmotions (Simplified) dataset — a multi-label emotion classification task with 28 emotion categories.

It uses sigmoid + BCEWithLogitsLoss and supports multiple simultaneous emotion predictions per sentence.

---

🏷️ Labels

There are 28 possible emotions, including:

admiration, amusement, anger, annoyance, approval, caring, confusion, curiosity, desire, disappointment, disapproval, embarrassment, excitement, fear, gratitude, grief, joy, love, nervousness, optimism, pride, realization, relief, remorse, sadness, surprise, neutral

---

📦 Training Setup

• Base model: distilbert-base-uncased
• Dataset: go_emotions (simplified version with 28 labels)
• Task type: Multi-label classification
• Loss: BCEWithLogitsLoss (enabled via problem_type="multi_label_classification")
• Epochs: 3
• Batch size: 16
• Learning rate: 2e-5
• Framework: transformers Trainer

---

🚀 Usage Example

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

model = AutoModelForSequenceClassification.from_pretrained("TuhinG/distilbert-goemotions")
tokenizer = AutoTokenizer.from_pretrained("TuhinG/distilbert-goemotions")

text = "I'm feeling excited and a little nervous about tomorrow."
inputs = tokenizer(text, return_tensors="pt")

with torch.no_grad():
    logits = model(**inputs).logits
    probs = torch.sigmoid(logits)[0]

# Threshold and map to labels
emotions = [
    "admiration", "amusement", "anger", "annoyance", "approval", "caring", "confusion", "curiosity",
    "desire", "disappointment", "disapproval", "embarrassment", "excitement", "fear", "gratitude", "grief",
    "joy", "love", "nervousness", "optimism", "pride", "realization", "relief", "remorse", "sadness",
    "surprise", "neutral"
]

for i, prob in enumerate(probs):
    if prob > 0.5:
        print(f"{emotions[i]}: {prob:.2f}")

---

📊 Evaluation Metric

During training, micro and macro F1 scores were used:

• f1_micro: high recall on rare labels
• f1_macro: balanced performance across labels

---

✨ Author

Fine-tuned and maintained by TuhinG

---

🛡️ License

Apache 2.0 — free to use, share, and modify.

---

## ✅ Next Steps

You can now:
1. Copy this to a file: `README.md`
2. Place it inside your model folder (`./distilbert_emotion_model`)
3. Push it to Hugging Face again:

```python
from huggingface_hub import upload_folder

upload_folder(
    folder_path="./distilbert_emotion_model",
    repo_id="TuhinG/distilbert-goemotions",
    repo_type="model",
    commit_message="Add model card README"
)

Let me know if you'd like me to generate the Python code for pushing this, or if you'd like to preview the model page before making it public.