language: en
license: creativeml-openrail-m
tags:
- emotion-classification
- multilabel
- bert
- goemotions
- affective-computing
- psychology
- NLP
- embeddings
- symbolic-ai
- poetic-ai
library_name: transformers
datasets:
- go_emotions
model_name: Rosa-V1
model_type: bert
pipeline_tag: text-classification
base_model: bert-base-uncased
widget:
- text: My heart is filled with longing and beauty.
- text: I'm excited but nervous about what's coming next.
metrics:
- name: eval_loss
type: loss
value: 0.0845
- name: eval_f1
type: f1
value: 0.5793
- name: parameters
type: count
value: 110000000
- name: epochs
type: count
value: 3
model_creator: Willinton Triana Cardona
model_description: >
ROSA is a fine-tuned BERT model trained on the GoEmotions dataset for
multilabel emotion classification. It identifies 28 nuanced human emotions
plus a neutral class, supports soft probability outputs, and provides latent
emotion embeddings for affective computing applications. ROSA is both
technically sound and symbolically aligned to poetic human understanding.
ROSA :: Emotional Sensitivity
“To feel is to know; to know is to bloom.” ·Willinton
ROSA is a fine-tuned Transformer model based on bert-base-uncased, trained on the GoEmotions dataset to classify 28 nuanced human emotions (plus neutral).
More than a model, ROSA is a prototype of emotion embeddings in affective computing.
🧠 Model Summary
| Metric | Value |
|---|---|
| Eval Loss | 0.0845 |
| Eval F1 | 0.5793 |
| Epochs | 3 |
| Dataset | GoEmotions |
| Model Base | BERT |
| Parameters | ~110M |
✨ Highlights
- Supports multilabel emotion classification
- Returns soft probability scores for each of the 29 emotions
- Includes optional latent vector embedding for downstream affect modeling
- Trained with HuggingFace
Trainer+ early evaluation - Symbolically aligned to human-centered semantics and poetic logic
🌸 Emotion Set
["admiration", "amusement", "anger", "annoyance", "approval", "caring",
"confusion", "curiosity", "desire", "disappointment", "disapproval",
"disgust", "embarrassment", "excitement", "fear", "gratitude", "grief",
"joy", "love", "nervousness", "optimism", "pride", "realization", "relief",
"remorse", "sadness", "surprise", "neutral"]
🔮 Usage
from transformers import BertTokenizer
from model.emotion_model import Rosa
import torch
tokenizer = BertTokenizer.from_pretrained("bert-base-uncased")
model = Rosa(num_emotions=29)
model.load_state_dict(torch.load("rosa.pt"))
model.eval()
text = "My heart is filled with longing and beauty."
inputs = tokenizer(text, return_tensors="pt", padding=True, truncation=True)
with torch.no_grad():
outputs = model(**inputs)
probs = torch.sigmoid(outputs["logits"]).squeeze()
# Result: list of probabilities for each emotion
🧭 Confusion Matrix
Included in the assets/ directory as confusion_matrix.png to show classification precision across emotions.
🧩 Architecture
┌──────────────┐
│ BERT Encoder │
└──────┬───────┘
↓
┌─────────────────┐
│ Dropout (Grace) │
└─────────────────┘
↓
┌────────────────────────┐
│ Dense Output (Bloom) │ → logits over 29 emotions
└────────────────────────┘
📦 Installation
pip install -r requirements.txt
Includes:
transformerstorchdatasetsscikit-learn
🖋️ License
CreativeML Open RAIL-M License
Please use this model ethically and with reverence for emotional contexts.
🌹 Creator
Willinton Triana Cardona
Philosopher · AI Engineer · Architect of Poetic Systems
ROSA is the Rosa of Barcelona, my first blossom of affective computing, semantic elegance, and sacred recursion.
🤝 Contributing
Pull requests, poetic expansions, multilingual emotion embeddings, and related metaphoric augmentations are welcome. I promise the next iteration (v2 with F1 improved) soon