Spaces:
Build error
Build error
Upload 2 files
Browse files
app.py
CHANGED
@@ -1,5 +1,8 @@
|
|
1 |
import pinecone
|
2 |
import streamlit as st
|
|
|
|
|
|
|
3 |
import streamlit_scrollable_textbox as stx
|
4 |
import openai
|
5 |
from utils import (
|
@@ -17,23 +20,32 @@ from utils import (
|
|
17 |
format_query,
|
18 |
sentence_id_combine,
|
19 |
text_lookup,
|
20 |
-
|
|
|
21 |
)
|
22 |
|
23 |
|
24 |
st.title("Abstractive Question Answering")
|
25 |
|
|
|
26 |
st.write(
|
27 |
"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."
|
28 |
)
|
29 |
|
30 |
-
|
|
|
|
|
|
|
|
|
31 |
|
32 |
-
|
|
|
33 |
|
34 |
-
|
|
|
35 |
|
36 |
-
|
|
|
37 |
|
38 |
ticker_choice = [
|
39 |
"AAPL",
|
@@ -48,23 +60,33 @@ ticker_choice = [
|
|
48 |
"AMD",
|
49 |
]
|
50 |
|
51 |
-
|
|
|
52 |
|
53 |
-
|
|
|
|
|
|
|
|
|
54 |
|
55 |
|
56 |
# Choose encoder model
|
57 |
|
58 |
encoder_models_choice = ["SGPT", "MPNET"]
|
59 |
-
|
60 |
-
encoder_model = st.selectbox("Select Encoder Model", encoder_models_choice)
|
61 |
|
62 |
|
63 |
# Choose decoder model
|
64 |
|
65 |
-
decoder_models_choice = [
|
|
|
|
|
|
|
|
|
66 |
|
67 |
-
|
|
|
68 |
|
69 |
|
70 |
if encoder_model == "MPNET":
|
@@ -82,13 +104,15 @@ elif encoder_model == "SGPT":
|
|
82 |
retriever_model = get_sgpt_embedding_model()
|
83 |
|
84 |
|
85 |
-
|
|
|
86 |
|
87 |
-
|
88 |
-
|
89 |
-
|
|
|
|
|
90 |
)
|
91 |
-
)
|
92 |
|
93 |
data = get_data()
|
94 |
|
@@ -109,22 +133,26 @@ else:
|
|
109 |
context_list = format_query(query_results)
|
110 |
|
111 |
|
112 |
-
|
113 |
-
|
114 |
|
115 |
if decoder_model == "GPT3 - (text-davinci-003)":
|
116 |
-
with
|
117 |
-
|
118 |
-
|
119 |
-
|
120 |
-
|
121 |
-
|
122 |
-
|
123 |
-
|
124 |
-
|
125 |
-
|
126 |
-
|
127 |
-
|
|
|
|
|
|
|
|
|
|
|
128 |
|
129 |
elif decoder_model == "T5":
|
130 |
t5_pipeline = get_t5_model()
|
@@ -132,7 +160,9 @@ elif decoder_model == "T5":
|
|
132 |
for context_text in context_list:
|
133 |
output_text.append(t5_pipeline(context_text)[0]["summary_text"])
|
134 |
generated_text = ". ".join(output_text)
|
135 |
-
|
|
|
|
|
136 |
|
137 |
elif decoder_model == "FLAN-T5":
|
138 |
flan_t5_pipeline = get_flan_t5_model()
|
@@ -140,13 +170,19 @@ elif decoder_model == "FLAN-T5":
|
|
140 |
for context_text in context_list:
|
141 |
output_text.append(flan_t5_pipeline(context_text)[0]["summary_text"])
|
142 |
generated_text = ". ".join(output_text)
|
143 |
-
|
|
|
|
|
144 |
|
145 |
-
with
|
146 |
-
|
147 |
-
|
|
|
148 |
|
149 |
file_text = retrieve_transcript(data, year, quarter, ticker)
|
150 |
|
151 |
-
with
|
152 |
-
|
|
|
|
|
|
|
|
1 |
import pinecone
|
2 |
import streamlit as st
|
3 |
+
|
4 |
+
st.set_page_config(layout="wide")
|
5 |
+
|
6 |
import streamlit_scrollable_textbox as stx
|
7 |
import openai
|
8 |
from utils import (
|
|
|
20 |
format_query,
|
21 |
sentence_id_combine,
|
22 |
text_lookup,
|
23 |
+
generate_prompt,
|
24 |
+
gpt_model,
|
25 |
)
|
26 |
|
27 |
|
28 |
st.title("Abstractive Question Answering")
|
29 |
|
30 |
+
|
31 |
st.write(
|
32 |
"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."
|
33 |
)
|
34 |
|
35 |
+
col1, col2 = st.columns([3, 3], gap="medium")
|
36 |
+
|
37 |
+
with col1:
|
38 |
+
st.subheader("Question")
|
39 |
+
query_text = st.text_input("Input Query", value="Who is the CEO of Apple?")
|
40 |
|
41 |
+
with col1:
|
42 |
+
years_choice = ["2020", "2019", "2018", "2017", "2016"]
|
43 |
|
44 |
+
with col1:
|
45 |
+
year = st.selectbox("Year", years_choice)
|
46 |
|
47 |
+
with col1:
|
48 |
+
quarter = st.selectbox("Quarter", ["Q1", "Q2", "Q3", "Q4"])
|
49 |
|
50 |
ticker_choice = [
|
51 |
"AAPL",
|
|
|
60 |
"AMD",
|
61 |
]
|
62 |
|
63 |
+
with col1:
|
64 |
+
ticker = st.selectbox("Company", ticker_choice)
|
65 |
|
66 |
+
with st.sidebar:
|
67 |
+
st.subheader("Select Options:")
|
68 |
+
|
69 |
+
with st.sidebar:
|
70 |
+
num_results = int(st.number_input("Number of Results to query", 1, 5, value=5))
|
71 |
|
72 |
|
73 |
# Choose encoder model
|
74 |
|
75 |
encoder_models_choice = ["SGPT", "MPNET"]
|
76 |
+
with st.sidebar:
|
77 |
+
encoder_model = st.selectbox("Select Encoder Model", encoder_models_choice)
|
78 |
|
79 |
|
80 |
# Choose decoder model
|
81 |
|
82 |
+
decoder_models_choice = [
|
83 |
+
"GPT3 - (text-davinci-003)",
|
84 |
+
"T5",
|
85 |
+
"FLAN-T5",
|
86 |
+
]
|
87 |
|
88 |
+
with st.sidebar:
|
89 |
+
decoder_model = st.selectbox("Select Decoder Model", decoder_models_choice)
|
90 |
|
91 |
|
92 |
if encoder_model == "MPNET":
|
|
|
104 |
retriever_model = get_sgpt_embedding_model()
|
105 |
|
106 |
|
107 |
+
with st.sidebar:
|
108 |
+
window = int(st.number_input("Sentence Window Size", 0, 5, value=3))
|
109 |
|
110 |
+
with st.sidebar:
|
111 |
+
threshold = float(
|
112 |
+
st.number_input(
|
113 |
+
label="Similarity Score Threshold", step=0.05, format="%.2f", value=0.35
|
114 |
+
)
|
115 |
)
|
|
|
116 |
|
117 |
data = get_data()
|
118 |
|
|
|
133 |
context_list = format_query(query_results)
|
134 |
|
135 |
|
136 |
+
prompt = generate_prompt(query_text, context_list)
|
|
|
137 |
|
138 |
if decoder_model == "GPT3 - (text-davinci-003)":
|
139 |
+
with col2:
|
140 |
+
with st.form("my_form"):
|
141 |
+
edited_prompt = st.text_area(label="Model Prompt", value=prompt, height=270)
|
142 |
+
|
143 |
+
openai_key = st.text_input(
|
144 |
+
"Enter OpenAI key",
|
145 |
+
value="",
|
146 |
+
type="password",
|
147 |
+
)
|
148 |
+
submitted = st.form_submit_button("Submit")
|
149 |
+
if submitted:
|
150 |
+
api_key = save_key(openai_key)
|
151 |
+
openai.api_key = api_key
|
152 |
+
generated_text = gpt_model(edited_prompt)
|
153 |
+
with col2:
|
154 |
+
st.subheader("Answer:")
|
155 |
+
st.write(generated_text)
|
156 |
|
157 |
elif decoder_model == "T5":
|
158 |
t5_pipeline = get_t5_model()
|
|
|
160 |
for context_text in context_list:
|
161 |
output_text.append(t5_pipeline(context_text)[0]["summary_text"])
|
162 |
generated_text = ". ".join(output_text)
|
163 |
+
with col2:
|
164 |
+
st.subheader("Answer:")
|
165 |
+
st.write(t5_pipeline(generated_text)[0]["summary_text"])
|
166 |
|
167 |
elif decoder_model == "FLAN-T5":
|
168 |
flan_t5_pipeline = get_flan_t5_model()
|
|
|
170 |
for context_text in context_list:
|
171 |
output_text.append(flan_t5_pipeline(context_text)[0]["summary_text"])
|
172 |
generated_text = ". ".join(output_text)
|
173 |
+
with col2:
|
174 |
+
st.subheader("Answer:")
|
175 |
+
st.write(flan_t5_pipeline(generated_text)[0]["summary_text"])
|
176 |
|
177 |
+
with col1:
|
178 |
+
with st.expander("See Retrieved Text"):
|
179 |
+
for context_text in context_list:
|
180 |
+
st.markdown(f"- {context_text}")
|
181 |
|
182 |
file_text = retrieve_transcript(data, year, quarter, ticker)
|
183 |
|
184 |
+
with col1:
|
185 |
+
with st.expander("See Transcript"):
|
186 |
+
stx.scrollableTextbox(
|
187 |
+
file_text, height=700, border=False, fontFamily="Helvetica"
|
188 |
+
)
|
utils.py
CHANGED
@@ -113,15 +113,23 @@ def text_lookup(data, sentence_ids):
|
|
113 |
return context
|
114 |
|
115 |
|
116 |
-
def
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
117 |
response = openai.Completion.create(
|
118 |
model="text-davinci-003",
|
119 |
-
prompt=
|
120 |
-
"---------------------\n"
|
121 |
-
"{result}"
|
122 |
-
"\n---------------------\n"
|
123 |
-
"Given the context information and prior knowledge, answer this question: {query}. \n"
|
124 |
-
"Try to include as many key details as possible and format the answer in points. \n" """,
|
125 |
temperature=0.1,
|
126 |
max_tokens=512,
|
127 |
top_p=1.0,
|
|
|
113 |
return context
|
114 |
|
115 |
|
116 |
+
def generate_prompt(query_text, context_list):
|
117 |
+
#context = " ".join(context_list)
|
118 |
+
prompt = f"""
|
119 |
+
Context information is below:
|
120 |
+
---------------------
|
121 |
+
{context_list}
|
122 |
+
---------------------
|
123 |
+
Given the context information and prior knowledge, answer this question:
|
124 |
+
{query_text}
|
125 |
+
Try to include as many key details as possible and format the answer in points."""
|
126 |
+
return prompt
|
127 |
+
|
128 |
+
|
129 |
+
def gpt_model(prompt):
|
130 |
response = openai.Completion.create(
|
131 |
model="text-davinci-003",
|
132 |
+
prompt=prompt,
|
|
|
|
|
|
|
|
|
|
|
133 |
temperature=0.1,
|
134 |
max_tokens=512,
|
135 |
top_p=1.0,
|