The model is trained on Hugging Face ag_news. The original World class is filtered out; Sports, Business, and Sci/Tech are mapped to the target classes.

Preprocessing

Input text is lowercased and cleaned from noisy symbols before BERT tokenization. The tokenizer uses max-length truncation and special tokens [CLS] and [SEP].

Metrics

Eval accuracy: 0.9110526315789473
Eval loss: 0.23549748957157135

TensorBoard

Nucelios/news-ai-bert

News AI BERT

Labels

Data

Preprocessing

Metrics

TensorBoard

Nucelios/
news-ai-bert