Create app.py

app.py

@@ -0,0 +1,48 @@
+import gradio as gr
+import torch
+from transformers import pipeline, PegasusForConditionalGeneration, PegasusTokenizer
+
+# Q&A model
+qa_classifier = pipeline(
+    "question-answering",
+    model="deepset/roberta-base-squad2",
+    tokenizer="deepset/roberta-base-squad2"
+)
+
+# paraphrase model
+paraphrase_model_name = 'tuner007/pegasus_paraphrase'
+paraphrase_tokenizer = PegasusTokenizer.from_pretrained(paraphrase_model_name)
+paraphrase_model = PegasusForConditionalGeneration.from_pretrained(paraphrase_model_name)
+
+
+def find_answer_and_paraphrase(text_input, question):
+    # Find the answer
+    answer = qa_classifier(context=text_input, question=question)["answer"]
+    
+    # Find the sentence containing the answer
+    sentences = text_input.split(".")
+    sentence_with_answer = ""
+    for sentence in sentences:
+        if answer in sentence:
+            sentence_with_answer = sentence.strip() + "."
+            break
+    
+    # Paraphrase the sentence containing the answer
+    input_text = "paraphrase: " + sentence_with_answer
+    num_return_sequences = 1
+    num_beams = 5
+    batch = paraphrase_tokenizer([input_text], truncation=True, padding='longest', max_length=60, return_tensors="pt").to(torch_device)
+    translated = paraphrase_model.generate(**batch, max_length=60, num_beams=num_beams, num_return_sequences=num_return_sequences, temperature=1.5)
+    paraphrase_text = paraphrase_tokenizer.batch_decode(translated, skip_special_tokens=True)[0]
+
+    return {"Answer": answer, "Paraphrase": paraphrase_text}
+
+
+inputs = [
+    gr.inputs.Textbox(label="Text Input"),
+    gr.inputs.Textbox(label="Question")
+]
+outputs = gr.outputs.Dict(label="Answer and Paraphrase")
+
+iface = gr.Interface(fn=find_answer_and_paraphrase, inputs=inputs, outputs=outputs, title="Question Answering and Paraphrasing", description="Enter the text and question to find the answer and paraphrase the sentence containing the answer.")
+iface.launch()