LLaMA-3.2B Fine-tuned for Emotion Classification (SemEval 2025)
This model is a fine-tuned version of meta-llama/Llama-3.2-3B for emotion classification on English text. It has been trained on the SemEval 2025 dataset to detect the presence of five key emotions in text: anger, fear, joy, sadness, and surprise.
Model Details
Model Description
This model classifies emotions in English text. It outputs the probability of five emotions (anger, fear, joy, sadness, surprise) for each input sentence.
- Model type: Transformer-based (LLaMA architecture)
- Language(s): English
- Finetuned from model:
meta-llama/Llama-3.2-3B
Model Sources
Uses
Direct Use
This model can be used to classify the emotions in any English text. The model outputs probabilities for five emotions, with higher values indicating stronger predictions.
Out-of-Scope Use
This model is not suitable for non-English text or highly specific emotion subcategories.
Bias, Risks, and Limitations
This model may misclassify emotions in texts with sarcasm, irony, or cultural nuances not well represented in the training data. The model also has limitations when applied to texts outside of the SemEval dataset's domain.
How to Get Started with the Model
You can load the model using Hugging Face's transformers library:
from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch
# Load the model and tokenizer
model = AutoModelForSequenceClassification.from_pretrained("gsi-upm/llama-3b-emotion-classifier", num_labels=5)
tokenizer = AutoTokenizer.from_pretrained("gsi-upm/llama-3b-emotion-classifier")
# Input text
text = "I just got a promotion and I feel amazing!"
# Tokenize the input
inputs = tokenizer(text, return_tensors="pt", padding=True, truncation=True)
# Predict emotions
with torch.no_grad():
logits = model(**inputs).logits
probs = torch.sigmoid(logits)
# Display results
labels = ['anger', 'fear', 'joy', 'sadness', 'surprise']
results = {label: float(prob) for label, prob in zip(labels, probs[0])}
print(results)
Output Format
The model outputs a dictionary of emotion probabilities:
{
'anger': 0.02,
'fear': 0.10,
'joy': 0.91,
'sadness': 0.05,
'surprise': 0.23
}
Training Details
Training Data
The model was trained on the SemEval 2025 dataset, which consists of English text with labeled emotions.
- Labels: anger, fear, joy, sadness, surprise
Evaluation
The model was evaluated on a held-out test set from the SemEval 2025 dataset using accuracy, precision, recall, and F1-score for each emotion class.
Results
- Precision: 0.7713
- Recall: 0.6607
- Weighted F1-score: 0.7421
Citation
If you use this model, please cite the following:
BibTeX:
@misc{llama-emotion-classifier,
author = {GSI-UPM},
title = {LLaMA-3.2B Fine-tuned for Emotion Classification},
year = {2025},
publisher = {Hugging Face},
url = {https://huggingface.co/gsi-upm/llama-3b-emotion-classifier}
}
APA:
Adrián Maldonado Robles. (2025). LLaMA-3.2B Fine-tuned for Emotion Classification. Hugging Face. Available at: https://huggingface.co/gsi-upm/llama-3b-emotion-classifier
Model Card Contact
For questions or inquiries, please contact [[email protected]].
Framework versions
- PEFT: 0.14.0
Model tree for gsi-upm/llama-3b-emotion-classifier
Base model
meta-llama/Llama-3.2-3B