DiariCore Mood Classifier

XLM-RoBERTa-base fine-tuned for 5-class emotion detection on English, Filipino, and Taglish diary entries.

Classes: angry, anxious, happy, neutral, sad

Architecture: XLM-RoBERTa-base + custom classification head (Dropout โ†’ Linear โ†’ LayerNorm โ†’ GELU โ†’ Dropout โ†’ Linear)

Training: Focal Loss + Label Smoothing, SWA, Differential LR, Keyword hybrid layer

Usage

See the DiariCore notebook for full inference code including calibration multipliers and keyword assist layer.

Downloads last month
9
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Space using sseia/diari-core-mood 1