mobilebert-uncased-finetuned-squadv2 Open-source Question Answering Model - Ultra Practical for Handling Answerable and Unanswerable Questions

Mobilebert Uncased Finetuned Squadv2

Developed by mrm8488

A QA model fine-tuned based on the lightweight MobileBERT architecture, specifically optimized for the SQuAD v2 dataset, capable of handling both answerable and unanswerable questions.

Question Answering System

Transformers

English#Lightweight QA #Unanswerable Detection #Low-resource Fine-tuning

Downloads 68

Release Time : 3/2/2022

Model Overview

This model is a fine-tuned version of MobileBERT on the SQuAD v2 dataset for QA tasks, able to determine whether an answer exists in the passage and provide corresponding responses.

Model Features

Lightweight & Efficient

Compared to standard BERT models, MobileBERT is smaller in size and more computationally efficient, making it suitable for mobile deployment.

Unanswerable Question Detection

Specifically optimized for the SQuAD v2 dataset, capable of identifying cases where no supporting answer exists in the passage.

Fast Convergence

The fine-tuning process converges quickly, significantly reducing computational costs.

Model Capabilities

Reading Comprehension

Question Answering System

Text Understanding

Unanswerable Question Detection

Use Cases

Educational Technology

Automated Answering System

Helps students automatically retrieve answers to questions based on textbook content

Approximately 75% EM score accuracy

Customer Service

FAQ Auto-Response

Answers common customer questions based on knowledge base content

Can identify questions not covered in the knowledge base

🚀 MobileBERT + SQuAD v2 📱❓

This is a fine-tuned mobilebert-uncased model on the SQUAD v2.0 dataset for the Q&A downstream task.

🚀 Quick Start

Fast usage with pipelines:

from transformers import pipeline
QnA_pipeline = pipeline('question-answering', model='mrm8488/mobilebert-uncased-finetuned-squadv2')
QnA_pipeline({
    'context': 'A new strain of flu that has the potential to become a pandemic has been identified in China by scientists.',
    'question': 'Who did identified it ?'
    })
    
# Output: {'answer': 'scientists.', 'end': 106, 'score': 0.41531604528427124, 'start': 96}

✨ Features

Model Features

MobileBERT is a thin version of BERT_LARGE, while equipped with bottleneck structures and a carefully designed balance between self-attentions and feed-forward networks. The checkpoint used here is the original MobileBert Optimized Uncased English: (uncased_L-24_H-128_B-512_A-4_F-4_OPT) checkpoint. More about the model here

Dataset Features

SQuAD2.0 combines the 100,000 questions in SQuAD1.1 with over 50,000 unanswerable questions written adversarially by crowdworkers to look similar to answerable ones. To do well on SQuAD2.0, systems must not only answer questions when possible, but also determine when no answer is supported by the paragraph and abstain from answering.

📦 Installation

The model was trained on a Tesla P100 GPU and 25GB of RAM with the following command:

python transformers/examples/question-answering/run_squad.py \
  --model_type bert \
  --model_name_or_path 'google/mobilebert-uncased' \
  --do_eval \
  --do_train \
  --do_lower_case \
  --train_file '/content/dataset/train-v2.0.json' \
  --predict_file '/content/dataset/dev-v2.0.json' \
  --per_gpu_train_batch_size 16 \
  --learning_rate 3e-5 \
  --num_train_epochs 5 \
  --max_seq_length 384 \
  --doc_stride 128 \
  --output_dir '/content/output' \
  --overwrite_output_dir \
  --save_steps 1000 \
  --version_2_with_negative

It is important to say that this models converges much faster than other ones. So, it is also cheap to fine-tune.

📚 Documentation

Test set Results

Property	Details
EM	75.37
F1	78.48
Size	94 MB

Created by Manuel Romero/@mrm8488 | LinkedIn

Made with ♥ in Spain

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