glue-mrpc-bert-base-uncased

This model is a fine-tuned version of bert-base-uncased on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.7308
• Acc: 0.8186
• F1: 0.8695

Model description

BERT Base Uncased Model

Intended uses & limitations

Binary text classification. Detecting whether two sentences are paraphrases of each other.

Training and evaluation data

Microsoft Research Paraphrase Corpus Dataset from the GLUE benchmark. Each sample contains a pair of sentences labeled as equivalent (paraphrased) or not equivalent (not paraphrased).

Training Samples: 3668

Validation Samples: 408

Test Samples: 1725

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 64
• eval_batch_size: 64
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 5

Training results

Training Loss	Epoch	Step	Validation Loss	Acc	F1
0.5781	1.0	58	0.5202	0.7507	0.8067
0.3782	2.0	116	0.4702	0.8	0.8606
0.1973	3.0	174	0.4731	0.8191	0.8659
0.1013	4.0	232	0.6764	0.8145	0.8662
0.0544	5.0	290	0.7308	0.8186	0.8695

Framework versions

• Transformers 4.48.3
• Pytorch 2.5.1+cu124
• Datasets 3.3.2
• Tokenizers 0.21.0