sweetpapa/moodsense-mood

MoodSense Mood Detection Model

Fine-tuned model for real-time mood detection from speech transcripts.

Model Details

• Base Model: google/gemma-3-1b-it
• Format: GGUF (Q4_K_M quantization)
• Size: ~518 MB
• Use Case: Emotion detection from text with function calling

Supported Emotions

• joy
• sadness
• anger
• fear
• surprise
• disgust
• neutral

Usage

This model is designed to work with MoodSense, a privacy-first mood detection application.

With llama.cpp

./main -m moodsense-mood-q4-k-m.gguf -p "Analyze the emotion: I'm so happy today!"

With node-llama-cpp

import { getLlama } from 'node-llama-cpp';

const llama = await getLlama();
const model = await llama.loadModel({ modelPath: 'moodsense-mood-q4-k-m.gguf' });

Output Format

The model outputs function calls in the format:

{
  "name": "update_mood",
  "arguments": {
    "emotion": "joy",
    "intensity": 0.8,
    "sentiment": "positive"
  }
}

Training

Fine-tuned using QLoRA on curated emotion datasets including:

• GoEmotions (Reddit comments)
• DailyDialog
• EmpatheticDialogues
• Custom conversational data

License

Apache 2.0