Upload 2 files

app.py

@@ -1,5 +1,8 @@
 import pinecone
 import streamlit as st
+
+st.set_page_config(layout="wide")
+
 import streamlit_scrollable_textbox as stx
 import openai
 from utils import (
@@ -17,23 +20,32 @@ from utils import (
     format_query,
     sentence_id_combine,
     text_lookup,
-    gpt3,
+    generate_prompt,
+    gpt_model,
 )
 
 
 st.title("Abstractive Question Answering")
 
+
 st.write(
     "The app uses the quarterly earnings call transcripts for 10 companies (Apple, AMD, Amazon, Cisco, Google, Microsoft, Nvidia, ASML, Intel, Micron) for the years 2016 to 2020."
 )
 
-query_text = st.text_input("Input Query", value="Who is the CEO of Apple?")
+col1, col2 = st.columns([3, 3], gap="medium")
+
+with col1:
+    st.subheader("Question")
+    query_text = st.text_input("Input Query", value="Who is the CEO of Apple?")
 
-years_choice = ["2020", "2019", "2018", "2017", "2016"]
+with col1:
+    years_choice = ["2020", "2019", "2018", "2017", "2016"]
 
-year = st.selectbox("Year", years_choice)
+with col1:
+    year = st.selectbox("Year", years_choice)
 
-quarter = st.selectbox("Quarter", ["Q1", "Q2", "Q3", "Q4"])
+with col1:
+    quarter = st.selectbox("Quarter", ["Q1", "Q2", "Q3", "Q4"])
 
 ticker_choice = [
     "AAPL",
@@ -48,23 +60,33 @@ ticker_choice = [
     "AMD",
 ]
 
-ticker = st.selectbox("Company", ticker_choice)
+with col1:
+    ticker = st.selectbox("Company", ticker_choice)
 
-num_results = int(st.number_input("Number of Results to query", 1, 5, value=5))
+with st.sidebar:
+    st.subheader("Select Options:")
+
+with st.sidebar:
+    num_results = int(st.number_input("Number of Results to query", 1, 5, value=5))
 
 
 # Choose encoder model
 
 encoder_models_choice = ["SGPT", "MPNET"]
-
-encoder_model = st.selectbox("Select Encoder Model", encoder_models_choice)
+with st.sidebar:
+    encoder_model = st.selectbox("Select Encoder Model", encoder_models_choice)
 
 
 # Choose decoder model
 
-decoder_models_choice = ["FLAN-T5", "T5", "GPT3 - (text-davinci-003)"]
+decoder_models_choice = [
+    "GPT3 - (text-davinci-003)",
+    "T5",
+    "FLAN-T5",
+]
 
-decoder_model = st.selectbox("Select Decoder Model", decoder_models_choice)
+with st.sidebar:
+    decoder_model = st.selectbox("Select Decoder Model", decoder_models_choice)
 
 
 if encoder_model == "MPNET":
@@ -82,13 +104,15 @@ elif encoder_model == "SGPT":
     retriever_model = get_sgpt_embedding_model()
 
 
-window = int(st.number_input("Sentence Window Size", 0, 5, value=3))
+with st.sidebar:
+    window = int(st.number_input("Sentence Window Size", 0, 5, value=3))
 
-threshold = float(
-    st.number_input(
-        label="Similarity Score Threshold", step=0.05, format="%.2f", value=0.35
+with st.sidebar:
+    threshold = float(
+        st.number_input(
+            label="Similarity Score Threshold", step=0.05, format="%.2f", value=0.35
+        )
     )
-)
 
 data = get_data()
 
@@ -109,22 +133,26 @@ else:
     context_list = format_query(query_results)
 
 
-st.subheader("Answer:")
-
+prompt = generate_prompt(query_text, context_list)
 
 if decoder_model == "GPT3 - (text-davinci-003)":
-    with st.form("my_form"):
-        openai_key = st.text_input(
-            "Enter OpenAI key",
-            value="",
-            type="password",
-        )
-        submitted = st.form_submit_button("Submit")
-        if submitted:
-            api_key = save_key(openai_key)
-            openai.api_key = api_key
-            generated_text = gpt3(query_text, context_list)
-            st.write(generated_text)
+    with col2:
+        with st.form("my_form"):
+            edited_prompt = st.text_area(label="Model Prompt", value=prompt, height=270)
+
+            openai_key = st.text_input(
+                "Enter OpenAI key",
+                value="",
+                type="password",
+            )
+            submitted = st.form_submit_button("Submit")
+            if submitted:
+                api_key = save_key(openai_key)
+                openai.api_key = api_key
+                generated_text = gpt_model(edited_prompt)
+                with col2:
+                    st.subheader("Answer:")
+                    st.write(generated_text)
 
 elif decoder_model == "T5":
     t5_pipeline = get_t5_model()
@@ -132,7 +160,9 @@ elif decoder_model == "T5":
     for context_text in context_list:
         output_text.append(t5_pipeline(context_text)[0]["summary_text"])
     generated_text = ". ".join(output_text)
-    st.write(t5_pipeline(generated_text)[0]["summary_text"])
+    with col2:
+        st.subheader("Answer:")
+        st.write(t5_pipeline(generated_text)[0]["summary_text"])
 
 elif decoder_model == "FLAN-T5":
     flan_t5_pipeline = get_flan_t5_model()
@@ -140,13 +170,19 @@ elif decoder_model == "FLAN-T5":
     for context_text in context_list:
         output_text.append(flan_t5_pipeline(context_text)[0]["summary_text"])
     generated_text = ". ".join(output_text)
-    st.write(flan_t5_pipeline(generated_text)[0]["summary_text"])
+    with col2:
+        st.subheader("Answer:")
+        st.write(flan_t5_pipeline(generated_text)[0]["summary_text"])
 
-with st.expander("See Retrieved Text"):
-    for context_text in context_list:
-        st.markdown(f"- {context_text}")
+with col1:
+    with st.expander("See Retrieved Text"):
+        for context_text in context_list:
+            st.markdown(f"- {context_text}")
 
 file_text = retrieve_transcript(data, year, quarter, ticker)
 
-with st.expander("See Transcript"):
-    stx.scrollableTextbox(file_text, height=700, border=False, fontFamily="Helvetica")
+with col1:
+    with st.expander("See Transcript"):
+        stx.scrollableTextbox(
+            file_text, height=700, border=False, fontFamily="Helvetica"
+        )

utils.py

@@ -113,15 +113,23 @@ def text_lookup(data, sentence_ids):
     return context
 
 
-def gpt3(query, result):
+def generate_prompt(query_text, context_list):
+    #context = " ".join(context_list)
+    prompt = f"""
+    Context information is below: 
+    ---------------------
+    {context_list}
+    ---------------------
+    Given the context information and prior knowledge, answer this question:
+    {query_text} 
+    Try to include as many key details as possible and format the answer in points."""
+    return prompt
+
+
+def gpt_model(prompt):
     response = openai.Completion.create(
         model="text-davinci-003",
-        prompt=f"""Context information is below. \n"
-    "---------------------\n"
-    "{result}"
-    "\n---------------------\n"
-    "Given the context information and prior knowledge, answer this question: {query}. \n"
-    "Try to include as many key details as possible and format the answer in points. \n" """,
+        prompt=prompt,
         temperature=0.1,
         max_tokens=512,
         top_p=1.0,