Fine tuning · Evaluation · Inference
One story. Four categories.
Classify a news headline or short article with your fine-tuned DistilBERT model. Select a registered version, then compare predictions as the model learns from more examples.
Try a news story
Choose an example or enter your own text.
0 / 4,000 characters
Model predictions
Category scores and GPU inference time for each selected version.
Your results will appear here
Select a model and classify a story.
Compare two versions on the same text.
Scores show the model's relative confidence across four categories. The model reads up to 128 tokens from your text. The displayed inference time excludes model loading and network travel.
01 / TEST
Establish the starting pointTest registered version 1 and inspect category errors on the fixed held-out examples.02 / RETRAIN
Learn from more examplesTrain from the original pretrained model with 32,000 examples on the separate training A10.03 / COMPARE
Measure the changeUse identical test examples, then compare registered versions on your own news text.