LaBSE Attention Pooling Polarization Classifier
This model is based on LaBSE with attention pooling and multi-layer concatenation for polarization classification.
Model Architecture
- Base Model: setu4993/LaBSE
- Pooling: Attention pooling (instead of mean pooling)
- Layers: Last 6 layers pooled and concatenated
- Classifier: 3 hidden layers with sizes [512, 256, 128]
Training
- Languages: 22 languages
- Max Length: 128
- Batch Size: 64
- Learning Rate: 2e-05
- Downloads last month
- -