sjrhuschlee's picture
Update README.md
caa89c2
|
raw
history blame
5.42 kB
metadata
license: mit
datasets:
  - squad_v2
  - squad
  - mrqa
  - mbartolo/synQA
  - adversarial_qa
  - newsqa
  - trivia_qa
  - search_qa
  - hotpot_qa
  - natural_questions
language:
  - en
library_name: transformers
pipeline_tag: question-answering
tags:
  - deberta
  - deberta-v3
  - question-answering
  - squad
  - squad_v2
  - mrqa
  - synQA
  - adversarial_qa
base_model: microsoft/deberta-v3-base
model-index:
  - name: sjrhuschlee/deberta-v3-base-squad2-ext-v1
    results:
      - task:
          type: question-answering
          name: Question Answering
        dataset:
          name: squad_v2
          type: squad_v2
          config: squad_v2
          split: validation
        metrics:
          - type: exact_match
            value: 79.483
            name: Exact Match
          - type: f1
            value: 82.343
            name: F1
      - task:
          type: question-answering
          name: Question Answering
        dataset:
          name: squad
          type: squad
          config: plain_text
          split: validation
        metrics:
          - type: exact_match
            value: 87.985
            name: Exact Match
          - type: f1
            value: 93.651
            name: F1
      - task:
          type: question-answering
          name: Question Answering
        dataset:
          name: adversarial_qa
          type: adversarial_qa
          config: adversarialQA
          split: validation
        metrics:
          - type: exact_match
            value: 47.533
            name: Exact Match
          - type: f1
            value: 59.838
            name: F1
      - task:
          type: question-answering
          name: Question Answering
        dataset:
          name: squad_adversarial
          type: squad_adversarial
          config: AddOneSent
          split: validation
        metrics:
          - type: exact_match
            value: 84.723
            name: Exact Match
          - type: f1
            value: 89.78
            name: F1
      - task:
          type: question-answering
          name: Question Answering
        dataset:
          name: squadshifts amazon
          type: squadshifts
          config: amazon
          split: test
        metrics:
          - type: exact_match
            value: 74.851
            name: Exact Match
          - type: f1
            value: 87.448
            name: F1
      - task:
          type: question-answering
          name: Question Answering
        dataset:
          name: squadshifts new_wiki
          type: squadshifts
          config: new_wiki
          split: test
        metrics:
          - type: exact_match
            value: 83.396
            name: Exact Match
          - type: f1
            value: 91.996
            name: F1
      - task:
          type: question-answering
          name: Question Answering
        dataset:
          name: squadshifts nyt
          type: squadshifts
          config: nyt
          split: test
        metrics:
          - type: exact_match
            value: 83.934
            name: Exact Match
          - type: f1
            value: 92.234
            name: F1
      - task:
          type: question-answering
          name: Question Answering
        dataset:
          name: squadshifts reddit
          type: squadshifts
          config: reddit
          split: test
        metrics:
          - type: exact_match
            value: 75.008
            name: Exact Match
          - type: f1
            value: 86.12
            name: F1

deberta-v3-base for Extractive QA

This is the deberta-v3-base model, fine-tuned using the SQuAD 2.0, MRQA, AdversarialQA, and SynQA datasets. It's been trained on question-answer pairs, including unanswerable questions, for the task of Extractive Question Answering.

Overview

Language model: deberta-v3-base
Language: English
Downstream-task: Extractive QA
Training data: SQuAD 2.0, MRQA, AdversarialQA, SynQA
Eval data: SQuAD 2.0
Infrastructure: 1x NVIDIA 3070

Model Usage

import torch
from transformers import(
  AutoModelForQuestionAnswering,
  AutoTokenizer,
  pipeline
)
model_name = "sjrhuschlee/deberta-v3-base-squad2-ext-v1"

# a) Using pipelines
nlp = pipeline('question-answering', model=model_name, tokenizer=model_name)
qa_input = {
'question': 'Where do I live?',
'context': 'My name is Sarah and I live in London'
}
res = nlp(qa_input)
# {'score': 0.984, 'start': 30, 'end': 37, 'answer': ' London'}

# b) Load model & tokenizer
model = AutoModelForQuestionAnswering.from_pretrained(model_name)
tokenizer = AutoTokenizer.from_pretrained(model_name)

question = 'Where do I live?'
context = 'My name is Sarah and I live in London'
encoding = tokenizer(question, context, return_tensors="pt")
start_scores, end_scores = model(
  encoding["input_ids"],
  attention_mask=encoding["attention_mask"],
  return_dict=False
)

all_tokens = tokenizer.convert_ids_to_tokens(input_ids[0].tolist())
answer_tokens = all_tokens[torch.argmax(start_scores):torch.argmax(end_scores) + 1]
answer = tokenizer.decode(tokenizer.convert_tokens_to_ids(answer_tokens))
# 'London'

Dataset Preparation

The MRQA dataset was updated to fix some errors and formatting to work with the run_qa.py example script provided in the Hugging Face Transformers library. The changes included

  • Updating incorrect answer starts locations (usually off by a few characters)
  • Updating the answer text to exactly match the text found in the context

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

  • learning_rate: 1e-06
  • train_batch_size: 12
  • eval_batch_size: 8
  • seed: 42
  • gradient_accumulation_steps: 8
  • total_train_batch_size: 96
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_ratio: 0.1
  • num_epochs: 3.0

Framework versions

  • Transformers 4.31.0.dev0