DistilBERT-base Fine-Tuned on AG News (Full Dataset)

Fine-tuned distilbert-base-uncased on the full AG News dataset (120,000 training examples, 4 classes) for use as the evaluation checkpoint in the A3 compression sweep.

  • Base Model: distilbert-base-uncased
  • Training Samples: 120,000
  • Validation Samples: 7,600
  • Epochs: 2
  • Batch Size: 128 (FP16)
  • Validation Accuracy: 94.26%
  • Macro F1: 0.9428
Downloads last month
4
Safetensors
Model size
67M params
Tensor type
F32
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Dataset used to train Hooshaai/distilbert-agnews

Evaluation results