This is the deberta-v3-large model, fine-tuned using the SQuAD2.0 dataset. It's been trained on question-answer pairs, including unanswerable questions, for the task of Extractive Question Answering.

This model was trained using LoRA available through the PEFT library.

Overview

Language model: deberta-v3-large
Language: English
Downstream-task: Extractive QA
Training data: SQuAD 2.0
Eval data: SQuAD 2.0
Infrastructure: 1x NVIDIA 3070

Model Usage

Using Transformers

This uses the merged weights (base model weights + LoRA weights) to allow for simple use in Transformers pipelines. It has the same performance as using the weights separately when using the PEFT library.

python

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

# 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'

Metrics

bash

# Squad v2
{
    "eval_HasAns_exact": 84.83468286099865,
    "eval_HasAns_f1": 90.48374860633226,
    "eval_HasAns_total": 5928,
    "eval_NoAns_exact": 91.0681244743482,
    "eval_NoAns_f1": 91.0681244743482,
    "eval_NoAns_total": 5945,
    "eval_best_exact": 87.95586625115808,
    "eval_best_exact_thresh": 0.0,
    "eval_best_f1": 90.77635490089573,
    "eval_best_f1_thresh": 0.0,
    "eval_exact": 87.95586625115808,
    "eval_f1": 90.77635490089592,
    "eval_runtime": 623.1333,
    "eval_samples": 11951,
    "eval_samples_per_second": 19.179,
    "eval_steps_per_second": 0.799,
    "eval_total": 11873
}

# Squad
{
    "eval_exact_match": 89.29044465468307,
    "eval_f1": 94.9846365606959,
    "eval_runtime": 553.7132,
    "eval_samples": 10618,
    "eval_samples_per_second": 19.176,
    "eval_steps_per_second": 0.8
}

Using with Peft

NOTE: This requires code in the PR https://github.com/huggingface/peft/pull/473 for the PEFT library.

python

#!pip install peft

from peft import LoraConfig, PeftModelForQuestionAnswering
from transformers import AutoModelForQuestionAnswering, AutoTokenizer
model_name = "sjrhuschlee/deberta-v3-large-squad2"

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 5e-05
train_batch_size: 24
eval_batch_size: 8
seed: 42
gradient_accumulation_steps: 1
total_train_batch_size: 24
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
lr_scheduler_warmup_ratio: 0.1
num_epochs: 4.0

LoRA Config

{
  "base_model_name_or_path": "microsoft/deberta-v3-large",
  "bias": "none",
  "fan_in_fan_out": false,
  "inference_mode": true,
  "init_lora_weights": true,
  "lora_alpha": 32,
  "lora_dropout": 0.1,
  "modules_to_save": ["qa_outputs"],
  "peft_type": "LORA",
  "r": 8,
  "target_modules": [
    "query_proj",
    "key_proj",
    "value_proj",
    "dense"
  ],
  "task_type": "QUESTION_ANS"
}

Framework versions

Transformers 4.30.0.dev0
Pytorch 2.0.1+cu117
Datasets 2.12.0
Tokenizers 0.13.3