BERT Base
Bidirectional transformer for text classification tasks
BERT Base Uncased is a bidirectional transformer pre-trained on large text corpora. At inference it takes a text string and returns predicted class labels and their probabilities. Requires a fine-tuned checkpoint.
When to use:
- Sentiment analysis (positive / negative / neutral)
- Topic or intent classification
- Spam detection or content moderation
Input: Text string to classify + optional fine-tuned checkpoint Output: Predicted class labels (CSV) and class probabilities per prediction
Model Settings
Max Seq Length (default: 512) Maximum number of tokens the model processes. Texts longer than this are truncated.
- 128: Fast, suitable for short sentences
- 256: Good for paragraphs
- 512: Full document understanding (default, slower)
Inference Settings
No dedicated inference-time settings. The model classifies text deterministically using the loaded checkpoint.