Modelpublic

Fatpandac/emojitrans

sourceHugging Facemitupdated 4mo agoView on Hugging Face

0likes4downloads

Model Card

emojitrans

A fine-tuned multilingual emoji prediction model based on microsoft/Multilingual-MiniLM-L12-H384.

Given a text input, the model predicts the top-K most relevant emojis (32-class classification).

—Base model: Multilingual-MiniLM-L12-H384 (12-layer Transformer, hidden size 384)
—Training data: vincentclaes/emoji-predictor — 14,420 tweets
—Val Top-5 accuracy: 53%
—Runtime: ONNX (quantized), runs in-browser via onnxruntime-web

File	Description
`model_q8.onnx`	INT8 quantized ONNX model (~120 MB)
`tokenizer.json`	Tokenizer vocab & rules
`tokenizer_config.json`	Tokenizer config
`config.json`	Model architecture config
`id2label.json`	Label index → emoji character mapping