Update app.py

app.py

@@ -0,0 +1,69 @@
+import gradio as gr
+import pysbd
+from transformers import pipeline
+from sentence_transformers import CrossEncoder
+from  transformers  import  AutoTokenizer, AutoModelWithLMHead, pipeline
+
+model_name = "MaRiOrOsSi/t5-base-finetuned-question-answering"
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+model = AutoModelWithLMHead.from_pretrained(model_name)
+
+#from transformers import pipeline
+
+#text2text_generator = pipeline("text2text-generation", model = "gpt2")
+
+sentence_segmenter = pysbd.Segmenter(language='en',clean=False)
+passage_retreival_model = CrossEncoder('cross-encoder/ms-marco-MiniLM-L-6-v2')
+qa_model = pipeline("question-answering",'a-ware/bart-squadv2')
+
+def fetch_answers(question, clincal_note ):
+    clincal_note_paragraphs = clincal_note.splitlines()
+    query_paragraph_list = [(question, para) for para in clincal_note_paragraphs if len(para.strip()) > 0 ]
+    
+    scores = passage_retreival_model.predict(query_paragraph_list)
+    top_5_indices = scores.argsort()[-5:]
+    top_5_query_paragraph_list = [query_paragraph_list[i] for i in top_5_indices ]
+    top_5_query_paragraph_list.reverse()
+    
+    top_5_query_paragraph_answer_list = ""
+    count = 1
+    for query, passage in top_5_query_paragraph_list:
+     passage_sentences = sentence_segmenter.segment(passage)
+     answer = qa_model(question = query, context = passage)['answer']
+     evidence_sentence = ""
+     for i in range(len(passage_sentences)):
+         if answer.startswith('.') or answer.startswith(':'):
+             answer = answer[1:].strip()
+         if answer in passage_sentences[i]:
+             evidence_sentence = evidence_sentence + " " + passage_sentences[i]
+     
+                  
+     model_input = f"question: {query} context: {evidence_sentence}"
+     #output_answer = text2text_generator(model_input)[0]['generated_text']
+     encoded_input = tokenizer([model_input],
+                               return_tensors='pt',
+                               max_length=512,
+                               truncation=True)
+                                     
+     output = model.generate(input_ids = encoded_input.input_ids,
+                             attention_mask = encoded_input.attention_mask)
+     output_answer = tokenizer.decode(output[0], skip_special_tokens=True)
+
+     result_str = "# ANSWER "+str(count)+": "+ output_answer +"\n"  
+     result_str = result_str + "REFERENCE: "+ evidence_sentence + "\n\n"
+     top_5_query_paragraph_answer_list += result_str
+     count+=1
+     
+    return top_5_query_paragraph_answer_list
+
+demo = gr.Interface(
+    fn=fetch_answers,
+    #take input as real time audio and use OPENAPI whisper for S2T
+    #clinical note upload as file (.This is an example of simple text. or doc/docx file)
+    inputs=[gr.Textbox(lines=2, label='Question', show_label=True),
+            gr.Textbox(lines=10, label='Document Text', show_label=True)],
+    outputs="markdown",
+    examples='.',
+    title='Document Question Answering System with Evidence from document'
+)
+demo.launch()