Avishkaarak-ekta for QA

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

Overview

Language model: avishkaarak-ekta-hindi
Language: English, Hindi(Upcoming)
Downstream-task: Extractive QA
Training data: SQuAD 2.0
Eval data: SQuAD 2.0
Code: See an example QA pipeline on Haystack
Infrastructure: 4x Tesla v100

Hyperparameters

batch_size = 4
n_epochs = 50
base_LM_model = "roberta-base"
max_seq_len = 512
learning_rate = 9e-5
lr_schedule = LinearWarmup
warmup_proportion = 0.2
doc_stride=128
max_query_length=64

Usage

In Haystack

Haystack is an NLP framework by deepset. You can use this model in a Haystack pipeline to do question answering at scale (over many documents). To load the model in Haystack:

reader = FARMReader(model_name_or_path="AVISHKAARAM/avishkaarak-ekta-hindi")
# or 
reader = TransformersReader(model_name_or_path="AVISHKAARAM/avishkaarak-ekta-hindi",tokenizer="AVISHKAARAM/avishkaarak-ekta-hindi")

For a complete example of AVISHKAARAM/avishkaarak-ekta-hindi being used for Question Answering, check out the Tutorials in Haystack Documentation

In Transformers

from transformers import AutoModelForQuestionAnswering, AutoTokenizer, pipeline

model_name = "AVISHKAARAM/avishkaarak-ekta-hindi"

# a) Get predictions
nlp = pipeline('question-answering', model=model_name, tokenizer=model_name)
QA_input = {
    'question': 'What is Bhagavadgita?',
    'context': 'Bhagavadgita, (Sanskrit: “Song of God”) an episode recorded in the great Sanskrit poem of the Hindus, the Mahabharata. It occupies chapters 23 to 40 of Book VI of the Mahabharata and is composed in the form of a dialogue between Prince Arjuna and Krishna, an avatar (incarnation) of the god Vishnu. Composed perhaps in the 1st or 2nd century CE, it is commonly known as the Gita.'
}
res = nlp(QA_input)

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

Performance

Evaluated on the SQuAD 2.0 dev set with the official eval script.

"exact": 79.87029394424324,
"f1": 82.91251169582613,

"total": 11873,
"HasAns_exact": 77.93522267206478,
"HasAns_f1": 84.02838248389763,
"HasAns_total": 5928,
"NoAns_exact": 81.79983179142137,
"NoAns_f1": 81.79983179142137,
"NoAns_total": 5945

Authors

Shashwat Bindal: avishkaaram.models@gmail.com

Sanoj: avishkaaram.models@gmail.com