Update app.py

T5 question generation

app.py

@@ -152,7 +152,8 @@
 # grad.Interface(classify, inputs=[txt,labels], outputs=out).launch()
 
 #-----------------------------------------------------------------------------------
-# 8. Text Generation Task/Models
+# 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.
 
@@ -205,17 +206,50 @@
 #-----------------------------------------------------------------------------------
 # 9. Text Generation: different model "distilgpt2"
 
-from transformers import pipeline, set_seed
+# 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 - first use case 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
 
-gpt2_pipe = pipeline('text-generation', model='distilgpt2')
-set_seed(42)
+text2text_tkn = AutoTokenizer.from_pretrained("mrm8488/t5-base-finetuned-question-generation-ap")
+mdl = AutoModelWithLMHead.from_pretrained("mrm8488/t5-base-finetuned-question-generation-ap")
 
-def generate(starting_text):
-    response= gpt2_pipe(starting_text, max_length=20, num_return_sequences=5)
+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
 
-txt=grad.Textbox(lines=1, label="English", placeholder="English Text here")
-out=grad.Textbox(lines=1, label="Generated Text")
+context=grad.Textbox(lines=10, label="English", placeholder="Context")
+ans=grad.Textbox(lines=1, label="Answer")
+out=grad.Textbox(lines=1, label="Genereated Question")
+
+grad.Interface(text2text, inputs=[context,ans], outputs=out).launch()
+
+
 
-grad.Interface(generate, inputs=txt, outputs=out).launch()