app.py

# 1. The RoBERTa base model is used, fine-tuned using the SQuAD 2.0 dataset. 
# It’s been trained on question-answer pairs, including unanswerable questions, for the task of question and answering.

# from transformers import AutoModelForQuestionAnswering, AutoTokenizer, pipeline
# import gradio as grad
# import ast

# mdl_name = "deepset/roberta-base-squad2"
# my_pipeline = pipeline('question-answering', model=mdl_name, tokenizer=mdl_name)

# def answer_question(question,context):
#     text= "{"+"'question': '"+question+"','context': '"+context+"'}"
#     di=ast.literal_eval(text)
#     response = my_pipeline(di)
#     return response

# grad.Interface(answer_question, inputs=["text","text"], outputs="text").launch()

#---------------------------------------------------------------------------------
# 2. Same task, different model.

# from transformers import AutoModelForQuestionAnswering, AutoTokenizer, pipeline
# import gradio as grad
# import ast

# mdl_name = "distilbert-base-cased-distilled-squad"
# my_pipeline = pipeline('question-answering', model=mdl_name, tokenizer=mdl_name)

# def answer_question(question,context):
#     text= "{"+"'question': '"+question+"','context': '"+context+"'}"
#     di=ast.literal_eval(text)
#     response = my_pipeline(di)
#     return response

# grad.Interface(answer_question, inputs=["text","text"], outputs="text").launch()

#---------------------------------------------------------------------------------
# 3. Different task: language translation.

# from transformers import pipeline
# import gradio as grad

# First model translates English to German.
# mdl_name = "Helsinki-NLP/opus-mt-en-de"
# opus_translator = pipeline("translation", model=mdl_name)

# def translate(text):
#     response = opus_translator(text)
#     return response

# grad.Interface(translate, inputs=["text",], outputs="text").launch()

#----------------------------------------------------------------------------------
# 4. Language translation without pipeline API.
# Second model translates English to French.

# from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
# import gradio as grad

# mdl_name = "Helsinki-NLP/opus-mt-en-fr"
# mdl = AutoModelForSeq2SeqLM.from_pretrained(mdl_name)
# my_tkn = AutoTokenizer.from_pretrained(mdl_name)

# def translate(text):
#     inputs = my_tkn(text, return_tensors="pt")
#     trans_output = mdl.generate(**inputs)
#     response = my_tkn.decode(trans_output[0], skip_special_tokens=True)
#     return response

# txt = grad.Textbox(lines=1, label="English", placeholder="English Text here")
# out = grad.Textbox(lines=1, label="French")
# grad.Interface(translate, inputs=txt, outputs=out).launch()

#-----------------------------------------------------------------------------------
# 5. Different task: abstractive summarization
# Abstractive summarization is more difficult than extractive summarization, 
# which pulls key sentences from a document and combines them to form a “summary.” 
# Because abstractive summarization involves paraphrasing words, it is also more time-consuming; 
# however, it has the potential to produce a more polished and coherent summary.

# from transformers import PegasusForConditionalGeneration, PegasusTokenizer
# import gradio as grad

# mdl_name = "google/pegasus-xsum"
# pegasus_tkn = PegasusTokenizer.from_pretrained(mdl_name)
# mdl = PegasusForConditionalGeneration.from_pretrained(mdl_name)

# def summarize(text):
#     tokens = pegasus_tkn(text, truncation=True, padding="longest", return_tensors="pt")
#     txt_summary = mdl.generate(**tokens)
#     response = pegasus_tkn.batch_decode(txt_summary, skip_special_tokens=True)
#     return response

# txt = grad.Textbox(lines=10, label="English", placeholder="English Text here")
# out = grad.Textbox(lines=10, label="Summary")

# grad.Interface(summarize, inputs=txt, outputs=out).launch()

#------------------------------------------------------------------------------------------
# 6. Same model with some tuning with some parameters: num_return_sequences=5, max_length=200, temperature=1.5, num_beams=10

# from transformers import PegasusForConditionalGeneration, PegasusTokenizer
# import gradio as grad

# mdl_name = "google/pegasus-xsum"
# pegasus_tkn = PegasusTokenizer.from_pretrained(mdl_name)
# mdl = PegasusForConditionalGeneration.from_pretrained(mdl_name)

# def summarize(text):
#     tokens = pegasus_tkn(text, truncation=True, padding="longest", return_tensors="pt")
#     translated_txt = mdl.generate(**tokens, num_return_sequences=5, max_length=200, temperature=1.5, num_beams=10)
#     response = pegasus_tkn.batch_decode(translated_txt, skip_special_tokens=True)
#     return response

# txt = grad.Textbox(lines=10, label="English", placeholder="English Text here")
# out = grad.Textbox(lines=10, label="Summary")

# grad.Interface(summarize, inputs=txt, outputs=out).launch()

#-----------------------------------------------------------------------------------
# 7. Zero-Shot Learning: 
# Zero-shot learning, as the name implies, is to use a pretrained model , trained on a certain set of data, 
# on a different set of data, which it has not seen during training. This would mean, as an example, to take 
# some model from huggingface that is trained on a certain dataset and use it for inference on examples it has never seen before.

# The transformers are where the zero-shot classification implementations are most frequently found by us. 
# There are more than 60 transformer models that function based on the zero-shot classification that are found in the huggingface library.

# When we discuss zero-shot text classification , there is one additional thing that springs to mind. 
# In the same vein as zero-shot classification is few-shot classification, which is very similar to zero-shot classification. 
# However, in contrast with zero-shot classification, few-shot classification makes use of very few labeled samples during the training process. 
# The implementation of the few-shot classification methods can be found in OpenAI, where the GPT3 classifier is a well-known example of a few-shot classifier.

# Deploying the following code works but comes with a warning: "No model was supplied, defaulted to facebook/bart-large-mnli and revision c626438 (https://huggingface.co/facebook/bart-large-mnli).
# Using a pipeline without specifying a model name and revision in production is not recommended."

# from transformers import pipeline
# import gradio as grad

# zero_shot_classifier = pipeline("zero-shot-classification")

# def classify(text,labels):
#     classifer_labels = labels.split(",")
#     #["software", "politics", "love", "movies", "emergency", "advertisment","sports"]
#     response = zero_shot_classifier(text,classifer_labels)
#     return response

# txt=grad.Textbox(lines=1, label="English", placeholder="text to be classified")
# labels=grad.Textbox(lines=1, label="Labels", placeholder="comma separated labels")
# out=grad.Textbox(lines=1, label="Classification")

# grad.Interface(classify, inputs=[txt,labels], outputs=out).launch()

#-----------------------------------------------------------------------------------
# 8. Text Generation Task/Models with GPT2 model

# The earliest text generation models were based on Markov chains . Markov chains are like a state machine wherein 
# using only the previous state, the next state is predicted. This is similar also to what we studied in bigrams.

# Post the Markov chains, recurrent neural networks (RNNs) , which were capable of retaining a greater context of the text, were introduced. 
# They are based on neural network architectures that are recurrent in nature. RNNs are able to retain a greater context of the text that was introduced. 
# Nevertheless, the amount of information that these kinds of networks are able to remember is constrained, and it is also difficult to train them, 
# which means that they are not effective at generating lengthy texts. To counter this issue with RNNs, LSTM architectures were evolved, 
# which could capture long-term dependencies in text. Finally, we came to transformers, whose decoder architecture became popular for generative models 
# used for generating text as an example.

# from transformers import GPT2LMHeadModel,GPT2Tokenizer
# import gradio as grad

# mdl = GPT2LMHeadModel.from_pretrained('gpt2')
# gpt2_tkn=GPT2Tokenizer.from_pretrained('gpt2')

# def generate(starting_text):
#     tkn_ids = gpt2_tkn.encode(starting_text, return_tensors = 'pt')
    
#     # When no specific parameter is specified, the model performs a greedy search to find the next word, which entails selecting the word from all of the
#     # alternatives that has the highest probability of being correct. This process is deterministic in nature, which means that resultant text is the same
#     # as before if we use the same parameters.

#     # The num_beams parameter does a beam search: it returns the sequences that have the highest probability, and then, when it comes time to
#     # choose, it picks the one that has the highest probability.

#     # The do_sample parameter select the next word at random from the probability distribution.

#     # The temperature parameter controls the level of greed that the generative model exhibits.
#     # If the temperature is low, the probabilities of sample classes other than the one with the highest log probability will be low. 
#     # As a result, the model will probably output the text that is most correct, but it will be rather monotonous and contain only a small amount of variation.
#     # If the temperature is high, the model has a greater chance of outputting different words than those with the highest probability. 
#     # The generated text will feature a greater variety of topics, but there is also an increased likelihood that it will generate nonsense text and 
#     # contain grammatical errors.

#     # With less temperature (1.5 --> 0.1), the output becomes less variational.
#     gpt2_tensors = mdl.generate(tkn_ids, max_length=100, no_repeat_ngram_size=True, num_beams=3, do_sample=True, temperature=0.1)
#     response=""
#     #response = gpt2_tensors
#     for i, x in enumerate(gpt2_tensors):
#         response=response+f"{i}: {gpt2_tkn.decode(x, skip_special_tokens=True)}" # Decode tensors into text
#     return gpt2_tensors, response

# txt=grad.Textbox(lines=1, label="English", placeholder="English Text here")
# out_tensors=grad.Textbox(lines=1, label="Generated Tensors")
# out_text=grad.Textbox(lines=1, label="Generated Text")

# grad.Interface(generate, inputs=txt, outputs=[out_tensors, out_text]).launch()

#-----------------------------------------------------------------------------------
# 9. Text Generation: different model "distilgpt2"

# from transformers import pipeline, set_seed
# import gradio as grad

# gpt2_pipe = pipeline('text-generation', model='distilgpt2')
# set_seed(42)

# def generate(starting_text):
#     response= gpt2_pipe(starting_text, max_length=20, num_return_sequences=5)
#     return response

# txt=grad.Textbox(lines=1, label="English", placeholder="English Text here")
# out=grad.Textbox(lines=1, label="Generated Text")

# grad.Interface(generate, inputs=txt, outputs=out).launch()

#-----------------------------------------------------------------------------------
# 10. Text-to-Text Generation using the T5 model - Task 1 generates a question given some context.

# A transformer-based architecture that takes a text-to-text approach is referred to as T5, which stands for Text-to-Text Transfer Transformer.

# In the text-to-text approach, we take a task like Q&A, classification, summarization, code generation, etc. and turn it into a problem, 
# which provides the model with some form of input and then teaches it to generate some form of target text. This makes it possible to apply 
# the same model, loss function, hyperparameters, and other settings to all of our varied sets of responsibilities.

# from transformers import AutoModelWithLMHead, AutoTokenizer
# import gradio as grad

# text2text_tkn = AutoTokenizer.from_pretrained("mrm8488/t5-base-finetuned-question-generation-ap")
# mdl = AutoModelWithLMHead.from_pretrained("mrm8488/t5-base-finetuned-question-generation-ap")

# def text2text(context,answer):
#     input_text = "answer: %s  context: %s </s>" % (answer, context)
#     features = text2text_tkn ([input_text], return_tensors='pt')
#     output = mdl.generate(input_ids=features['input_ids'],
#                           attention_mask=features['attention_mask'],
#                           max_length=64)
#     response=text2text_tkn.decode(output[0])
#     return response

# context=grad.Textbox(lines=10, label="English", placeholder="Context")
# ans=grad.Textbox(lines=1, label="Answer")
# out=grad.Textbox(lines=1, label="Generated Question")

# grad.Interface(text2text, inputs=[context,ans], outputs=out).launch()

#-----------------------------------------------------------------------------------
# 11. Text-to-Text Generation using the T5 model - Task 2 summarizes a paragraph of text.

# from transformers import AutoTokenizer, AutoModelWithLMHead
# import gradio as grad

# text2text_tkn = AutoTokenizer.from_pretrained("deep-learning-analytics/wikihow-t5-small")
# mdl = AutoModelWithLMHead.from_pretrained("deep-learning-analytics/wikihow-t5-small")

# def text2text_summary(para):
#     initial_txt = para.strip().replace("\n","")
#     tkn_text = text2text_tkn.encode(initial_txt, return_tensors="pt")
#     tkn_ids = mdl.generate(
#         tkn_text,
#         max_length=250,
#         num_beams=5,
#         repetition_penalty=2.5,
#         early_stopping=True
#     )
#     response = text2text_tkn.decode(tkn_ids[0], skip_special_tokens=True)
#     return response

# para=grad.Textbox(lines=10, label="Paragraph", placeholder="Copy paragraph")
# out=grad.Textbox(lines=1, label="Summary")

# grad.Interface(text2text_summary, inputs=para, outputs=out).launch()

#-----------------------------------------------------------------------------------
# 12. Text-to-Text Generation using the T5 model - Task 3 Translation.

# from transformers import T5ForConditionalGeneration, T5Tokenizer
# import gradio as grad

# text2text_tkn= T5Tokenizer.from_pretrained("t5-small")
# mdl = T5ForConditionalGeneration.from_pretrained("t5-small")

# def text2text_translation(text):
#     # English to German
#     # inp = "translate English to German:: "+text
#     # English to Frendh
#     inp = "translate English to French:: " +text
#     enc = text2text_tkn(inp, return_tensors="pt")
#     tokens = mdl.generate(**enc)
#     response=text2text_tkn.batch_decode(tokens)
#     return response

# para=grad.Textbox(lines=1, label="English Text", placeholder="Text in English")
# out=grad.Textbox(lines=1, label="French Translation")

# grad.Interface(text2text_translation, inputs=para, outputs=out).launch()

#-----------------------------------------------------------------------------------
# 13. Text-to-Text Generation using the T5 model - Task 4 sentiment analysis.

from transformers import T5ForConditionalGeneration, T5Tokenizer
import gradio as grad

text2text_tkn= T5Tokenizer.from_pretrained("t5-small")
mdl = T5ForConditionalGeneration.from_pretrained("t5-small")

def text2text_sentiment(text):
    inp = "sst2 sentence: "+text
    enc = text2text_tkn(inp, return_tensors="pt")
    tokens = mdl.generate(**enc)
    response=text2text_tkn.batch_decode(tokens)
    return response

para=grad.Textbox(lines=1, label="English Text", placeholder="Text in English")
out=grad.Textbox(lines=1, label="Sentiment")

grad.Interface(text2text_sentiment, inputs=para, outputs=out).launch()