This model was fine-tuned from ModernBERT-base on the GoEmotions dataset for multi-label classification. It predicts emotional states in text, with a total of 28 possible labels. Each input text can have one or more associated labels, reflecting the multi-label nature of the task.

Try it out here.

Model Details

Base Model: ModernBERT-base
Fine-Tuning Dataset: GoEmotions
Number of Labels: 28
Problem Type: Multi-label classification
Language: English
License: MIT
Fine-Tuning Framework: Hugging Face Transformers

Example Usage

Here’s how to use the model with Hugging Face Transformers:

python

from transformers import pipeline
import torch

# Load the model
classifier = pipeline(
    "text-classification", 
    model="cirimus/modernbert-base-go-emotions",
    return_all_scores=True
)

text = "I am so happy and excited about this opportunity!"
predictions = classifier(text)

# Print top 5 detected emotions
sorted_preds = sorted(predictions[0], key=lambda x: x['score'], reverse=True)
top_5 = sorted_preds[:5]

print("\nTop 5 emotions detected:")
for pred in top_5:
    print(f"\t{pred['label']:10s} : {pred['score']:.3f}")

## Example output:
# Top 5 emotions detected:
#        excitement : 0.937
#        joy        : 0.915
#        desire     : 0.022
#        love       : 0.020
#        admiration : 0.017

How the Model Was Created

The model was fine-tuned for 3 epochs using the following hyperparameters:

Learning Rate: 2e-5
Batch Size: 16
Weight Decay: 0.01
Warmup Steps: 500
Optimizer: AdamW
Evaluation Metrics: Precision, Recall, F1 Score (weighted), Accuracy

Dataset

The GoEmotions dataset is a multi-label emotion classification dataset derived from Reddit comments. It contains 58,000 examples with 28 emotion labels (e.g., admiration, amusement, anger, etc.), and it is annotated for multi-label classification.

Evaluation Results

The model was evaluated on the test split of the GoEmotions dataset, using a threshold of 0.5 for binarizing predictions. The overall metrics were:

Standard Results:

Using the default threshold of 0.5.

Label	Accuracy	Precision	Recall	F1	MCC	Support	Threshold
macro avg	0.970	0.665	0.389	0.465	0.477	5427	0.5
admiration	0.945	0.737	0.627	0.677	0.650	504	0.5
amusement	0.980	0.794	0.803	0.798	0.788	264	0.5
anger	0.968	0.680	0.258	0.374	0.406	198	0.5
annoyance	0.940	0.468	0.159	0.238	0.249	320	0.5
approval	0.942	0.614	0.276	0.381	0.387	351	0.5
caring	0.976	0.524	0.244	0.333	0.347	135	0.5
confusion	0.975	0.625	0.294	0.400	0.418	153	0.5
curiosity	0.951	0.538	0.423	0.473	0.452	284	0.5
desire	0.987	0.604	0.349	0.443	0.453	83	0.5
disappointment	0.974	0.656	0.139	0.230	0.294	151	0.5
disapproval	0.950	0.494	0.292	0.367	0.356	267	0.5
disgust	0.980	0.674	0.252	0.367	0.405	123	0.5
embarrassment	0.995	0.857	0.324	0.471	0.526	37	0.5
excitement	0.984	0.692	0.262	0.380	0.420	103	0.5
fear	0.992	0.796	0.551	0.652	0.659	78	0.5
gratitude	0.990	0.957	0.892	0.924	0.919	352	0.5
grief	0.999	0.000	0.000	0.000	0.000	6	0.5
joy	0.978	0.652	0.571	0.609	0.600	161	0.5
love	0.982	0.792	0.798	0.795	0.786	238	0.5
nervousness	0.996	0.636	0.304	0.412	0.439	23	0.5
optimism	0.975	0.743	0.403	0.523	0.536	186	0.5
pride	0.998	0.857	0.375	0.522	0.566	16	0.5
realization	0.973	0.514	0.124	0.200	0.244	145	0.5
relief	0.998	1.000	0.091	0.167	0.301	11	0.5
remorse	0.992	0.594	0.732	0.656	0.656	56	0.5
sadness	0.979	0.759	0.385	0.511	0.532	156	0.5
surprise	0.978	0.649	0.340	0.447	0.460	141	0.5
neutral	0.794	0.715	0.623	0.666	0.520	1787	0.5

Optimal Results:

Using the best threshold for each label based on the training set (tuned on F1).

Label	Accuracy	Precision	Recall	F1	MCC	Support	Threshold
macro avg	0.967	0.568	0.531	0.541	0.526	5427	various
admiration	0.946	0.700	0.726	0.713	0.683	504	0.30
amusement	0.981	0.782	0.856	0.817	0.808	264	0.40
anger	0.963	0.490	0.510	0.500	0.481	198	0.20
annoyance	0.917	0.337	0.425	0.376	0.334	320	0.25
approval	0.922	0.411	0.473	0.440	0.399	351	0.25
caring	0.971	0.424	0.415	0.419	0.405	135	0.25
confusion	0.970	0.468	0.484	0.476	0.460	153	0.30
curiosity	0.947	0.493	0.630	0.553	0.530	284	0.35
desire	0.988	0.708	0.410	0.519	0.533	83	0.45
disappointment	0.963	0.321	0.291	0.306	0.287	151	0.25
disapproval	0.943	0.429	0.464	0.446	0.417	267	0.30
disgust	0.981	0.604	0.496	0.545	0.538	123	0.20
embarrassment	0.995	0.789	0.405	0.536	0.564	37	0.30
excitement	0.979	0.444	0.388	0.415	0.405	103	0.25
fear	0.991	0.693	0.667	0.680	0.675	78	0.30
gratitude	0.990	0.951	0.886	0.918	0.913	352	0.50
grief	0.999	0.500	0.500	0.500	0.499	6	0.20
joy	0.978	0.628	0.609	0.618	0.607	161	0.40
love	0.982	0.789	0.819	0.804	0.795	238	0.45
nervousness	0.995	0.375	0.391	0.383	0.380	23	0.25
optimism	0.970	0.558	0.597	0.577	0.561	186	0.15
pride	0.998	0.750	0.375	0.500	0.529	16	0.15
realization	0.968	0.326	0.200	0.248	0.240	145	0.25
relief	0.998	0.429	0.273	0.333	0.341	11	0.25
remorse	0.993	0.611	0.786	0.688	0.689	56	0.55
sadness	0.979	0.667	0.538	0.596	0.589	156	0.20
surprise	0.978	0.585	0.511	0.545	0.535	141	0.30
neutral	0.782	0.649	0.737	0.690	0.526	1787	0.40

Intended Use

The model is designed for emotion classification in English-language text, particularly in domains such as:

Social media sentiment analysis
Customer feedback evaluation
Behavioral or psychological research

Limitations and Biases

Data Bias: The dataset is based on Reddit comments, which may not generalize well to other domains or cultural contexts.
Underrepresented Classes: Certain labels like "grief" and "relief" have very few examples, leading to lower performance for those classes.
Ambiguity: Some training data contain annotation inconsistencies or ambiguities that may impact predictions.

Environmental Impact

Hardware Used: NVIDIA RTX4090
Training Time: <1 hour
Carbon Emissions: ~0.04 kg CO2 (calculated via ML CO2 Impact Calculator).

Citation

If you use this model, please cite it as follows:

bibtex

@inproceedings{JdFE2025b,
  title = {Emotion Detection with ModernBERT},
  author = {Enric Junqu\'e de Fortuny},
  year = {2025},
  howpublished = {\url{https://huggingface.co/cirimus/modernbert-base-go-emotions}},
}

cirimus/modernbert-base-go-emotions-2025-01

Overview