YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
SentiLog Android Models
Packaged ONNX models for the SentiLog Android app — a privacy-first journaling app with on-device 28-emotion sentiment analysis.
Models
| Language | File | F1 Macro | Top-1 | Top-3 | Native Data |
|---|---|---|---|---|---|
| German | goemotions_de.zip |
0.447 | 69% | 82% | GermEval (10K) |
| Japanese | goemotions_ja.zip |
0.396 | 56% | 74% | WRIME (35K) |
| Korean | goemotions_ko.zip |
0.568 | 68% | 81% | KOTE (50K) |
Each ZIP contains:
model_quantized.onnx— INT8 quantised xlm-roberta-base (~206 MB)tokenizer.json— HuggingFace Unigram tokenizer for AndroidHfTokenizer
Base Model
Fine-tuned from xlm-roberta-base on GoEmotions (28 emotions) plus native-language datasets.
Version Manifest
models.json in this repo root is consumed by the app's ModelUpdateWorker to notify users of model updates.
Usage
These models are loaded by the SentiLog Android app automatically. They are not intended for standalone inference — use the PyTorch checkpoints on tojohere/goemotions-de-xlm-roberta-base, tojohere/goemotions-ja-xlm-roberta-base, and tojohere/goemotions-ko-xlm-roberta-base for research.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support