ModernBERT-base fine-tuned on BANKING77 (ONNX, fp16)
This is ModernBERT-base fine-tuned on BANKING77 to classify banking-support questions into 77 intents. It is part of the Accelerated Computational Linguistics (FA26) course project by Lihan Li and Xizi Luo.
The file onnx/model_fp16.onnx runs in the browser with transformers.js. On the official test set it scores 93.38% accuracy and 0.934 macro-F1, the same as the PyTorch checkpoint. Its argmax agrees with PyTorch on 99.9% of test queries.
Training: bf16, max_len 128, effective batch 32, AdamW (lr 8e-5, weight decay 0.01), 3 epochs, seed 42.