DistilBERT - MRPC Open-Source Text Classification Model: Freely Deployable for Judging Semantic Equivalence of Sentence Pairs

Distilbert Mrpc

Developed by mattchurgin

A text classification model fine-tuned on the GLUE dataset based on distilbert-base-uncased, used to determine whether sentence pairs are semantically equivalent

Text Classification

Transformers

Open Source License:Apache-2.0 #Text similarity judgment #Lightweight BERT #Sentence pair classification

Downloads 15

Release Time : 3/2/2022

Model Overview

This model is a fine-tuned version of DistilBERT, specifically designed for the MRPC (Microsoft Research Paraphrase Corpus) task, which determines whether two sentences express the same meaning.

Model Features

Efficient and Lightweight

Based on the DistilBERT architecture, it is 40% smaller than the full BERT model but retains 97% of its performance.

High Accuracy

Achieves 84.8% accuracy and 89.35% F1 score on the MRPC task.

Fast Inference

The distilled model design enables faster inference compared to the original BERT.

Model Capabilities

Text classification

Semantic similarity judgment

Sentence pair analysis

Use Cases

Text processing

QA system deduplication

Identify similar questions to avoid duplicate answers

Improves QA system efficiency

Content moderation

Detect duplicate or highly similar inappropriate content

Enhances moderation efficiency

Educational technology

Student answer grading

Determine the semantic similarity between student answers and standard answers

Assists automatic grading systems

🚀 distilbert-mrpc

This model is a fine - tuned version of [distilbert - base - uncased](https://huggingface.co/distilbert - base - uncased) on the glue dataset, achieving high accuracy and F1 scores on the evaluation set.

🚀 Quick Start

This model is a fine - tuned version of [distilbert - base - uncased](https://huggingface.co/distilbert - base - uncased) on the glue dataset. It achieves the following results on the evaluation set:

Loss: 0.6783
Accuracy: 0.8480
F1: 0.8935

📚 Documentation

Model Information

Property	Details
Model Type	Fine - tuned version of distilbert - base - uncased
Training Data	glue dataset
Metrics	Accuracy: 0.8480392156862745, F1: 0.8934707903780068

Model Index

Name: distilbert - mrpc
Results:
- Task:
  - Name: Text Classification
  - Type: text - classification
- Dataset:
  - Name: glue
  - Type: glue
  - Args: mrpc
- Metrics:
  - Name: Accuracy
  - Type: accuracy
  - Value: 0.8480392156862745
  - Name: F1
  - Type: f1
  - Value: 0.8934707903780068

🔧 Technical Details

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 5e - 05
train_batch_size: 8
eval_batch_size: 8
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon = 1e - 08
lr_scheduler_type: linear
num_epochs: 3.0

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1
0.5916	0.22	100	0.5676	0.7157	0.8034
0.5229	0.44	200	0.4534	0.7770	0.8212
0.5055	0.65	300	0.4037	0.8137	0.8762
0.4597	0.87	400	0.3706	0.8407	0.8893
0.4	1.09	500	0.4590	0.8113	0.8566
0.3498	1.31	600	0.4196	0.8554	0.8974
0.2916	1.53	700	0.4606	0.8554	0.8933
0.3309	1.74	800	0.5162	0.8578	0.9027
0.3788	1.96	900	0.3911	0.8529	0.8980
0.2059	2.18	1000	0.5842	0.8554	0.8995
0.1595	2.4	1100	0.5701	0.8578	0.8975
0.1205	2.61	1200	0.6905	0.8407	0.8889
0.174	2.83	1300	0.6783	0.8480	0.8935

Framework versions

Transformers 4.15.0
Pytorch 1.10.1
Datasets 1.17.0
Tokenizers 0.10.3

📄 License

This model is licensed under the Apache - 2.0 license.

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご