distilbert-base-uncased Quora Duplicate Pair Detection
This model is fine tuned of version of distilbert-base-uncased on quora dataset for detecting duplicate sentences or questions.
Loss :: 0.111300
Accuracy :: 0.900740
F1 Score :: 0.868633
Model Description:
DistilBERT model is a distilled form of the BERT model. The size of a BERT model was reduced by 40% via knowledge distillation during the pre-training phase while retaining 97% of its language understanding abilities and being 60% faster.
Training and evaluation data:
"quora"
Training Hyper Parameters:
learning_rate = 3e-4,
per_device_train_batch_size=32,
per_device_eval_batch_size=32,
num_train_epochs=4,
evaluation_strategy="epoch",
seed: 42,
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
Training Results:
Epoch Training Loss Validation Loss Accuracy F1
1 0.271500 0.264808 0.884909 0.844402
2 0.191200 0.258109 0.896399 0.866099
3 0.111300 0.315554 0.900740 0.868633
Label 0 = Not Duplicate
Label 1 = Duplicate
