Create app.py

initial commit

app.py

@@ -0,0 +1,60 @@
+import openai
+import numpy as np
+import pandas as pd
+from sentence_transformers import SentenceTransformer
+import re
+import pdftotext
+import gradio as gr
+import json
+
+# Calculate the cosine similarity
+def cos_sim(vector1, vector2):
+   cosine_similarity = np.dot(vector1, vector2) / (np.linalg.norm(vector1) * np.linalg.norm(vector2))
+   return cosine_similarity
+
+def sim_search(df, query, n=3, dot=False):
+   embedding = model.encode(query)
+   if dot: 
+      df['similarities'] = df.embeddings.apply(lambda x: x@embedding)
+      print("using dot product")
+   else:
+      df['similarities'] = df.embeddings.apply(lambda x: cos_sim(x, embedding))
+      print("using cosine similarity")
+   res = df.sort_values('similarities', ascending=False).head(n)
+   return res
+
+def create_prompt(context, question):
+    return f"""
+    Context information is below.
+    ---------------------
+    {context}
+    ---------------------
+    Given the context information and not prior knowledge, answer the query.
+    Query: {question}
+    Answer: \
+    """
+
+
+def answer_question(question, model="gpt-3.5-turbo",n=3):
+    r = sim_search(df, question,n=n)
+    context = "\n\n".join(r.chunks)
+    prompt = create_prompt(context, question)
+    response = openai.ChatCompletion.create(
+        model="gpt-3.5-turbo",
+        messages=[
+            {"role": "system", "content": "You are a helpful assistant answering questions in german. You answer only in german. If you do not know an answer you say it. You do not fabricate answers."},
+            {"role": "user", "content": prompt},
+        ]
+    )
+    return response.choices[0].message.content
+
+df = pd.read_csv("stroke_embeddings_sentence_transformers.csv")
+df["embeddings"] = df.embeddings.apply(json.loads)
+model = SentenceTransformer('thenlper/gte-base')
+
+def gradio_answer(input):
+    return answer_question(input)
+
+demo = gr.Interface(fn=gradio_answer, inputs=gr.Textbox(lines=1, placeholder="Frage hier...", label="Frage"), outputs=gr.Textbox(lines=4, placeholder="Antwort hier...", label="Antwort"), title="S3 Leitlinie Carotis Stenose", examples=["In welchen Intervallen ist eine Nachuntersuchung nach CAS angezeigt?", "Ist eine ambulante Therapie der Carotisstenose mittels CEA oder CAS möglich und sinnvoll?", "Was sollte man als Bradykardie-Therapie bei Nachdilatation eines Stents einsetzen?"])
+    
+demo.launch()