import gradio as gr
import torch
from transformers import (
pipeline,
AutoModelForSeq2SeqLM,
AutoTokenizer
)
M0 = "consciousAI/question-generation-auto-t5-v1-base-s"
M1 = "consciousAI/question-generation-auto-t5-v1-base-s-q"
M2 = "consciousAI/question-generation-auto-t5-v1-base-s-q-c"
M4 = "consciousAI/question-generation-auto-hints-t5-v1-base-s-q"
M5 = "consciousAI/question-generation-auto-hints-t5-v1-base-s-q-c"
device = ['cuda' if torch.cuda.is_available() else 'cpu'][0]
_m0 = AutoModelForSeq2SeqLM.from_pretrained(M0).to(device)
_tk0 = AutoTokenizer.from_pretrained(M0, cache_dir="./cache")
_m1 = AutoModelForSeq2SeqLM.from_pretrained(M1).to(device)
_tk1 = AutoTokenizer.from_pretrained(M1, cache_dir="./cache")
_m2 = AutoModelForSeq2SeqLM.from_pretrained(M2).to(device)
_tk2 = AutoTokenizer.from_pretrained(M2, cache_dir="./cache")
_m4 = AutoModelForSeq2SeqLM.from_pretrained(M4).to(device)
_tk4 = AutoTokenizer.from_pretrained(M4, cache_dir="./cache")
_m5 = AutoModelForSeq2SeqLM.from_pretrained(M5).to(device)
_tk5 = AutoTokenizer.from_pretrained(M5, cache_dir="./cache")
def _formatQs(questions):
_finalQs = ""
if questions is not None:
_qList = questions[0].strip().split("?")
qIdx = 1
if len(_qList) > 1:
for idx, _q in enumerate(_qList):
_q = _q.strip()
if _q is not None and len(_q) !=0:
_finalQs += str(qIdx) + ". " + _q + "? \n"
qIdx+=1
else:
if len(_qList[0])>1:
_finalQs = "1. " + str(_qList[0]) + "?"
else:
_finalQs = None
return _finalQs
def _generate(mode, context, hint=None, minLength=50, maxLength=500, lengthPenalty=2.0, earlyStopping=True, numReturnSequences=1, numBeams=2, noRepeatNGramSize=0, doSample=False, topK=0, penaltyAlpha=0, topP=0, temperature=0, model="All"):
predictionM0 = None
predictionM1 = None
predictionM2 = None
predictionM4 = None
predictionM5 = None
if mode == 'Auto':
_inputText = "question_context: " + context
if model == "All":
_encoding = _tk0.encode(_inputText, return_tensors='pt', truncation=True, padding='max_length').to(device) # max_length=1024
_outputEncoded = _m0.generate(_encoding,
min_length=minLength,
max_length=maxLength,
length_penalty=lengthPenalty,
early_stopping=earlyStopping,
num_return_sequences=numReturnSequences,
num_beams=numBeams,
no_repeat_ngram_size=noRepeatNGramSize,
do_sample=doSample,
top_k=topK,
penalty_alpha=penaltyAlpha,
top_p=topP,
temperature=temperature
)
predictionM0 = [_tk0.decode(id, clean_up_tokenization_spaces=False, skip_special_tokens=True) for id in _outputEncoded]
_encoding = _tk1.encode(_inputText, return_tensors='pt', truncation=True, padding='max_length').to(device) # max_length=1024
_outputEncoded = _m1.generate(_encoding,
min_length=minLength,
max_length=maxLength,
length_penalty=lengthPenalty,
early_stopping=earlyStopping,
num_return_sequences=numReturnSequences,
num_beams=numBeams,
no_repeat_ngram_size=noRepeatNGramSize,
do_sample=doSample,
top_k=topK,
penalty_alpha=penaltyAlpha,
top_p=topP,
temperature=temperature
)
predictionM1 = [_tk1.decode(id, clean_up_tokenization_spaces=False, skip_special_tokens=True) for id in _outputEncoded]
_encoding = _tk2.encode(_inputText, return_tensors='pt', truncation=True, padding='max_length').to(device) # max_length=1024 .to(device)
_outputEncoded = _m2.generate(_encoding,
min_length=minLength,
max_length=maxLength,
length_penalty=lengthPenalty,
early_stopping=earlyStopping,
num_return_sequences=numReturnSequences,
num_beams=numBeams,
no_repeat_ngram_size=noRepeatNGramSize,
do_sample=doSample,
top_k=topK,
penalty_alpha=penaltyAlpha,
top_p=topP,
temperature=temperature
)
predictionM2 = [_tk2.decode(id, clean_up_tokenization_spaces=False, skip_special_tokens=True) for id in _outputEncoded]
_encoding = _tk4.encode(_inputText, return_tensors='pt', truncation=True, padding='max_length').to(device) # max_length=1024 .to(device)
_outputEncoded = _m4.generate(_encoding,
min_length=minLength,
max_length=maxLength,
length_penalty=lengthPenalty,
early_stopping=earlyStopping,
num_return_sequences=numReturnSequences,
num_beams=numBeams,
no_repeat_ngram_size=noRepeatNGramSize,
do_sample=doSample,
top_k=topK,
penalty_alpha=penaltyAlpha,
top_p=topP,
temperature=temperature
)
predictionM4 = [_tk4.decode(id, clean_up_tokenization_spaces=False, skip_special_tokens=True) for id in _outputEncoded]
_encoding = _tk5.encode(_inputText, return_tensors='pt', truncation=True, padding='max_length').to(device) # max_length=1024 .to(device)
_outputEncoded = _m5.generate(_encoding,
min_length=minLength,
max_length=maxLength,
length_penalty=lengthPenalty,
early_stopping=earlyStopping,
num_return_sequences=numReturnSequences,
num_beams=numBeams,
no_repeat_ngram_size=noRepeatNGramSize,
do_sample=doSample,
top_k=topK,
penalty_alpha=penaltyAlpha,
top_p=topP,
temperature=temperature
)
predictionM5 = [_tk5.decode(id, clean_up_tokenization_spaces=False, skip_special_tokens=True) for id in _outputEncoded]
elif model == "question-generation-auto-hints-t5-v1-base-s-q-c":
_encoding = _tk5.encode(_inputText, return_tensors='pt', truncation=True, padding='max_length').to(device) # max_length=1024 .to(device)
_outputEncoded = _m5.generate(_encoding,
min_length=minLength,
max_length=maxLength,
length_penalty=lengthPenalty,
early_stopping=earlyStopping,
num_return_sequences=numReturnSequences,
num_beams=numBeams,
no_repeat_ngram_size=noRepeatNGramSize,
do_sample=doSample,
top_k=topK,
penalty_alpha=penaltyAlpha,
top_p=topP,
temperature=temperature
)
predictionM5 = [_tk5.decode(id, clean_up_tokenization_spaces=False, skip_special_tokens=True) for id in _outputEncoded]
elif model == "question-generation-auto-hints-t5-v1-base-s-q":
_encoding = _tk4.encode(_inputText, return_tensors='pt', truncation=True, padding='max_length').to(device) # max_length=1024 .to(device)
_outputEncoded = _m4.generate(_encoding,
min_length=minLength,
max_length=maxLength,
length_penalty=lengthPenalty,
early_stopping=earlyStopping,
num_return_sequences=numReturnSequences,
num_beams=numBeams,
no_repeat_ngram_size=noRepeatNGramSize,
do_sample=doSample,
top_k=topK,
penalty_alpha=penaltyAlpha,
top_p=topP,
temperature=temperature
)
predictionM4 = [_tk4.decode(id, clean_up_tokenization_spaces=False, skip_special_tokens=True) for id in _outputEncoded]
elif model == "question-generation-auto-t5-v1-base-s-q-c":
_encoding = _tk2.encode(_inputText, return_tensors='pt', truncation=True, padding='max_length').to(device) # max_length=1024 .to(device)
_outputEncoded = _m2.generate(_encoding,
min_length=minLength,
max_length=maxLength,
length_penalty=lengthPenalty,
early_stopping=earlyStopping,
num_return_sequences=numReturnSequences,
num_beams=numBeams,
no_repeat_ngram_size=noRepeatNGramSize,
do_sample=doSample,
top_k=topK,
penalty_alpha=penaltyAlpha,
top_p=topP,
temperature=temperature
)
predictionM2 = [_tk2.decode(id, clean_up_tokenization_spaces=False, skip_special_tokens=True) for id in _outputEncoded]
elif model == "question-generation-auto-t5-v1-base-s-q":
_encoding = _tk1.encode(_inputText, return_tensors='pt', truncation=True, padding='max_length').to(device) # max_length=1024
_outputEncoded = _m1.generate(_encoding,
min_length=minLength,
max_length=maxLength,
length_penalty=lengthPenalty,
early_stopping=earlyStopping,
num_return_sequences=numReturnSequences,
num_beams=numBeams,
no_repeat_ngram_size=noRepeatNGramSize,
do_sample=doSample,
top_k=topK,
penalty_alpha=penaltyAlpha,
top_p=topP,
temperature=temperature
)
predictionM1 = [_tk1.decode(id, clean_up_tokenization_spaces=False, skip_special_tokens=True) for id in _outputEncoded]
elif model == "question-generation-auto-t5-v1-base-s":
_encoding = _tk0.encode(_inputText, return_tensors='pt', truncation=True, padding='max_length').to(device) # max_length=1024
_outputEncoded = _m0.generate(_encoding,
min_length=minLength,
max_length=maxLength,
length_penalty=lengthPenalty,
early_stopping=earlyStopping,
num_return_sequences=numReturnSequences,
num_beams=numBeams,
no_repeat_ngram_size=noRepeatNGramSize,
do_sample=doSample,
top_k=topK,
penalty_alpha=penaltyAlpha,
top_p=topP,
temperature=temperature
)
predictionM0 = [_tk0.decode(id, clean_up_tokenization_spaces=False, skip_special_tokens=True) for id in _outputEncoded]
elif mode == 'Hints':
_inputText = "question_hint: " + hint + "</s>question_context: " + context
_encoding = _tk4.encode(_inputText, return_tensors='pt', truncation=True, padding='max_length').to(device) # max_length=1024 .to(device)
_outputEncoded = _m4.generate(_encoding,
min_length=minLength,
max_length=maxLength,
length_penalty=lengthPenalty,
early_stopping=earlyStopping,
num_return_sequences=numReturnSequences,
num_beams=numBeams,
no_repeat_ngram_size=noRepeatNGramSize,
do_sample=doSample,
top_k=topK,
penalty_alpha=penaltyAlpha,
top_p=topP,
temperature=temperature
)
predictionM4 = [_tk4.decode(id, clean_up_tokenization_spaces=False, skip_special_tokens=True) for id in _outputEncoded]
_encoding = _tk5.encode(_inputText, return_tensors='pt', truncation=True, padding='max_length').to(device) # max_length=1024 .to(device)
_outputEncoded = _m5.generate(_encoding,
min_length=minLength,
max_length=maxLength,
length_penalty=lengthPenalty,
early_stopping=earlyStopping,
num_return_sequences=numReturnSequences,
num_beams=numBeams,
no_repeat_ngram_size=noRepeatNGramSize,
do_sample=doSample,
top_k=topK,
penalty_alpha=penaltyAlpha,
top_p=topP,
temperature=temperature
)
predictionM5 = [_tk5.decode(id, clean_up_tokenization_spaces=False, skip_special_tokens=True) for id in _outputEncoded]
predictionM0 = _formatQs(predictionM0)
predictionM1 = _formatQs(predictionM1)
predictionM2 = _formatQs(predictionM2)
predictionM4 = _formatQs(predictionM4)
predictionM5 = _formatQs(predictionM5)
return predictionM5, predictionM4, predictionM2, predictionM1, predictionM0
with gr.Blocks() as demo:
gr.Markdown(value="# Question Generation Demo \n [question-generation-auto-t5-v1-base-s](https://huggingface.co/anshoomehra/question-generation-auto-t5-v1-base-s) ✫ [question-generation-auto-t5-v1-base-s-q](https://huggingface.co/anshoomehra/question-generation-auto-t5-v1-base-s-q) ✫ [question-generation-auto-t5-v1-base-s-q-c](https://huggingface.co/anshoomehra/question-generation-auto-t5-v1-base-s-q-c) ✫ [question-generation-auto-hints-t5-v1-base-s-q](https://huggingface.co/anshoomehra/question-generation-auto-hints-t5-v1-base-s-q) ✫ [question-generation-auto-hints-t5-v1-base-s-q-c](https://huggingface.co/anshoomehra/question-generation-auto-hints-t5-v1-base-s-q-c)\n\n Please be patient, 5 models may take up to 80 sec to run on CPU")
with gr.Accordion(variant='compact', label='Search Methods: Deteriminstic / Stochastic / Contrastive', open=True):
with gr.Row():
mode = gr.Radio(["Auto", "Hints"], value="Auto", label="Mode")
with gr.Row():
minLength = gr.Slider(10, 512, 50, step=1, label="Min Length")
maxLength = gr.Slider(20, 512, 164, step=1, label="Max Length")
lengthPenalty = gr.Slider(-5, 5, 1, label="Length Penalty")
earlyStopping = gr.Checkbox(True, label="Early Stopping [EOS]")
numReturnSequences = gr.Slider(1, 3, 1, step=1, label="Num return Sequences")
with gr.Row():
numBeams = gr.Slider(1, 10, 4, step=1, label="Beams")
noRepeatNGramSize = gr.Slider(0, 5, 3, step=1, label="No Repeat N-Gram Size")
with gr.Row():
doSample = gr.Checkbox(label="Do Random Sample")
topK = gr.Slider(0, 50, 0, step=1, label="Top K")
penaltyAlpha = gr.Slider(0.0, 1, 0, label="Penalty Alpha")
topP = gr.Slider(0, 1, 0, label="Top P/Nucleus Sampling")
temperature = gr.Slider(0.01, 1, 1, label="Temperature")
with gr.Row():
model = gr.Dropdown(["question-generation-auto-hints-t5-v1-base-s-q-c", "question-generation-auto-hints-t5-v1-base-s-q", "question-generation-auto-t5-v1-base-s-q-c", "question-generation-auto-t5-v1-base-s-q", "question-generation-auto-t5-v1-base-s", "All"], label="Model", value="question-generation-auto-hints-t5-v1-base-s-q-c")
with gr.Accordion(variant='compact', label='Input Values'):
with gr.Row(variant='compact'):
contextDefault = "Google LLC is an American multinational technology company focusing on search engine technology, online advertising, cloud computing, computer software, quantum computing, e-commerce, artificial intelligence, and consumer electronics. It has been referred to as 'the most powerful company in the world' and one of the world's most valuable brands due to its market dominance, data collection, and technological advantages in the area of artificial intelligence. Its parent company Alphabet is considered one of the Big Five American information technology companies, alongside Amazon, Apple, Meta, and Microsoft."
hintDefault = ""
context = gr.Textbox(contextDefault, label="Context", placeholder="Dummy Context", lines=5)
hint = gr.Textbox(hintDefault, label="Hint", placeholder="Enter hint here. Ensure the mode is set to 'Hints' prior using hints.", lines=2)
with gr.Accordion(variant='compact', label='Multi-Task Model(s) Sensitive To Hints'):
with gr.Row(variant='compact'):
_predictionM5 = gr.Textbox(label="Predicted Questions - question-generation-auto-hints-t5-v1-base-s-q-c [Hints Sensitive]")
_predictionM4 = gr.Textbox(label="Predicted Questions - question-generation-auto-hints-t5-v1-base-s-q [Hints Sensitive]")
with gr.Accordion(variant='compact', label='Uni-Task Model(s) Non-Sensitive To Hints'):
with gr.Row(variant='compact'):
_predictionM2 = gr.Textbox(label="Predicted Questions - question-generation-auto-t5-v1-base-s-q-c [No Hints]")
_predictionM1 = gr.Textbox(label="Predicted Questions - question-generation-auto-t5-v1-base-s-q [No Hints]")
_predictionM0 = gr.Textbox(label="Predicted Questions - question-generation-auto-t5-v1-base-s [No Hints]")
with gr.Row():
gen_btn = gr.Button("Generate Questions")
gen_btn.click(fn=_generate,
inputs=[mode, context, hint, minLength, maxLength, lengthPenalty, earlyStopping, numReturnSequences, numBeams, noRepeatNGramSize, doSample, topK, penaltyAlpha, topP, temperature, model],
outputs=[_predictionM5, _predictionM4, _predictionM2, _predictionM1, _predictionM0]
)
demo.launch(show_error=True)