metadata

tags:
  - Question(s) Generation
metrics:
  - rouge
model-index:
  - name: consciousAI/question-generation-auto-hints-t5-v1-base-s-q
    results: []

Auto Question Generation

The model is intended to be used for Auto And/Or Hint enabled Question Generation tasks. The model is expected to produce one or possibly more than one question from the provided context.

Live Demo: Question Generation

Including this there are five models trained with different training sets, demo provide comparison to all in one go. However, you can reach individual projects at below links:

Auto Question Generation v1

Auto Question Generation v2

Auto Question Generation v3

Auto/Hints based Question Generation v2

This model can be used as below:

from transformers import (
    AutoModelForSeq2SeqLM,
    AutoTokenizer
)

model_checkpoint = "consciousAI/question-generation-auto-hints-t5-v1-base-s-q"

model = AutoModelForSeq2SeqLM.from_pretrained(model_checkpoint)
tokenizer = AutoTokenizer.from_pretrained(model_checkpoint)

## Input with prompt
context="question_context: <context>"
encodings = tokenizer.encode(context, return_tensors='pt', truncation=True, padding='max_length').to(device)

## You can play with many hyperparams to condition the output, look at demo
output = model.generate(encodings, 
                        #max_length=300, 
                        #min_length=20, 
                        #length_penalty=2.0, 
                        num_beams=4,
                        #early_stopping=True,
                        #do_sample=True,
                        #temperature=1.1
                       )

## Multiple questions are expected to be delimited by '?' You can write a small wrapper to elegantly format. Look at the demo.
questions = [tokenizer.decode(id, clean_up_tokenization_spaces=False, skip_special_tokens=False) for id in output]

Training and evaluation data

Squad & QNLi combo.

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 0.0003
train_batch_size: 4
eval_batch_size: 4
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
num_epochs: 10

Training results

Training Loss	Epoch	Step	Validation Loss	Rouge1	Rouge2	Rougel	Rougelsum
1.8298	1.0	14515	1.7529	0.3535	0.1825	0.3251	0.3294
1.4931	2.0	29030	1.7132	0.3558	0.1881	0.3267	0.3308
1.2756	3.0	43545	1.7579	0.3604	0.1901	0.3307	0.3345
1.0936	4.0	58060	1.8173	0.36	0.1901	0.3295	0.3334
0.955	5.0	72575	1.9204	0.3611	0.1884	0.3295	0.3336
0.8117	6.0	87090	2.0183	0.355	0.1836	0.3241	0.3282
0.6949	7.0	101605	2.1347	0.3556	0.1836	0.3242	0.3282
0.636	8.0	116120	2.2567	0.3568	0.1855	0.3248	0.3286
0.591	9.0	130635	2.3598	0.3563	0.1844	0.3238	0.3281
0.5417	10.0	145150	2.4725	0.3556	0.1828	0.3229	0.3269

Framework versions

Transformers 4.23.0.dev0
Pytorch 1.12.1+cu113
Datasets 2.5.2
Tokenizers 0.13.0