Spaces:
Build error
Build error
Upload 8 files
Browse files- app.py +287 -129
- utils/models.py +8 -6
- utils/prompts.py +48 -1
- utils/retriever.py +52 -0
app.py
CHANGED
|
@@ -1,3 +1,5 @@
|
|
|
|
|
|
|
|
| 1 |
import openai
|
| 2 |
import streamlit_scrollable_textbox as stx
|
| 3 |
|
|
@@ -25,7 +27,7 @@ from utils.models import (
|
|
| 25 |
get_spacy_model,
|
| 26 |
get_splade_sparse_embedding_model,
|
| 27 |
get_t5_model,
|
| 28 |
-
|
| 29 |
save_key,
|
| 30 |
)
|
| 31 |
from utils.prompts import (
|
|
@@ -36,8 +38,10 @@ from utils.prompts import (
|
|
| 36 |
generate_flant5_prompt_summ_chunk_context_single,
|
| 37 |
generate_gpt_j_two_shot_prompt_1,
|
| 38 |
generate_gpt_j_two_shot_prompt_2,
|
| 39 |
-
|
| 40 |
-
|
|
|
|
|
|
|
| 41 |
get_context_list_prompt,
|
| 42 |
)
|
| 43 |
from utils.retriever import (
|
|
@@ -46,6 +50,7 @@ from utils.retriever import (
|
|
| 46 |
query_pinecone_sparse,
|
| 47 |
sentence_id_combine,
|
| 48 |
text_lookup,
|
|
|
|
| 49 |
)
|
| 50 |
from utils.transcript_retrieval import retrieve_transcript
|
| 51 |
from utils.vector_index import (
|
|
@@ -66,59 +71,29 @@ col1, col2 = st.columns([3, 3], gap="medium")
|
|
| 66 |
|
| 67 |
with st.sidebar:
|
| 68 |
ner_choice = st.selectbox("Select NER Model", ["Spacy", "Alpaca"])
|
|
|
|
|
|
|
|
|
|
| 69 |
|
| 70 |
if ner_choice == "Spacy":
|
| 71 |
ner_model = get_spacy_model()
|
| 72 |
|
| 73 |
with col1:
|
| 74 |
st.subheader("Question")
|
| 75 |
-
|
| 76 |
-
|
| 77 |
-
|
| 78 |
-
|
| 79 |
-
|
| 80 |
-
if ner_choice == "Alpaca":
|
| 81 |
-
ner_prompt = generate_alpaca_ner_prompt(query_text)
|
| 82 |
-
entity_text = generate_entities_flan_alpaca_inference_api(ner_prompt)
|
| 83 |
-
company_ent, quarter_ent, year_ent = format_entities_flan_alpaca(
|
| 84 |
-
entity_text
|
| 85 |
-
)
|
| 86 |
-
else:
|
| 87 |
-
company_ent = extract_ticker_spacy(query_text, ner_model)
|
| 88 |
-
quarter_ent, year_ent = extract_quarter_year(query_text)
|
| 89 |
-
|
| 90 |
-
ticker_index, quarter_index, year_index = clean_entities(
|
| 91 |
-
company_ent, quarter_ent, year_ent
|
| 92 |
-
)
|
| 93 |
-
|
| 94 |
-
with col1:
|
| 95 |
-
years_choice = ["2020", "2019", "2018", "2017", "2016", "All"]
|
| 96 |
-
|
| 97 |
-
with col1:
|
| 98 |
-
# Hardcoding the defaults for a question without metadata
|
| 99 |
-
if (
|
| 100 |
-
query_text
|
| 101 |
-
== "What was discussed regarding Wearables revenue performance?"
|
| 102 |
-
):
|
| 103 |
-
year = st.selectbox("Year", years_choice)
|
| 104 |
-
else:
|
| 105 |
-
year = st.selectbox("Year", years_choice, index=year_index)
|
| 106 |
-
|
| 107 |
-
with col1:
|
| 108 |
-
# Hardcoding the defaults for a question without metadata
|
| 109 |
-
if (
|
| 110 |
-
query_text
|
| 111 |
-
== "What was discussed regarding Wearables revenue performance?"
|
| 112 |
-
):
|
| 113 |
-
quarter = st.selectbox("Quarter", ["Q1", "Q2", "Q3", "Q4", "All"])
|
| 114 |
else:
|
| 115 |
-
|
| 116 |
-
"
|
|
|
|
| 117 |
)
|
| 118 |
|
| 119 |
-
with col1:
|
| 120 |
-
participant_type = st.selectbox("Speaker", ["Company Speaker", "Analyst"])
|
| 121 |
|
|
|
|
|
|
|
| 122 |
ticker_choice = [
|
| 123 |
"AAPL",
|
| 124 |
"CSCO",
|
|
@@ -132,23 +107,87 @@ ticker_choice = [
|
|
| 132 |
"AMD",
|
| 133 |
]
|
| 134 |
|
| 135 |
-
|
| 136 |
-
|
| 137 |
-
if
|
| 138 |
-
query_text
|
| 139 |
-
|
| 140 |
-
|
| 141 |
-
|
|
|
|
| 142 |
else:
|
| 143 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 144 |
|
| 145 |
with st.sidebar:
|
| 146 |
st.subheader("Select Options:")
|
| 147 |
|
| 148 |
-
|
| 149 |
-
|
| 150 |
-
|
| 151 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 152 |
|
| 153 |
|
| 154 |
# Choose encoder model
|
|
@@ -160,8 +199,11 @@ with st.sidebar:
|
|
| 160 |
|
| 161 |
# Choose decoder model
|
| 162 |
|
| 163 |
-
|
| 164 |
-
|
|
|
|
|
|
|
|
|
|
| 165 |
with st.sidebar:
|
| 166 |
decoder_model = st.selectbox("Select Decoder Model", decoder_models_choice)
|
| 167 |
|
|
@@ -198,66 +240,140 @@ elif encoder_model == "Hybrid MPNET - SPLADE":
|
|
| 198 |
) = get_splade_sparse_embedding_model()
|
| 199 |
|
| 200 |
with st.sidebar:
|
| 201 |
-
|
| 202 |
-
|
| 203 |
-
|
| 204 |
-
|
| 205 |
-
|
| 206 |
-
|
| 207 |
-
|
| 208 |
-
|
| 209 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 210 |
)
|
| 211 |
-
)
|
| 212 |
|
| 213 |
data = get_data()
|
| 214 |
|
| 215 |
-
if
|
| 216 |
-
|
| 217 |
-
|
| 218 |
-
|
| 219 |
-
|
| 220 |
-
|
| 221 |
-
|
| 222 |
-
|
| 223 |
-
dense_query_embedding, sparse_query_embedding
|
| 224 |
-
|
| 225 |
-
|
| 226 |
-
|
| 227 |
-
|
| 228 |
-
|
| 229 |
-
|
| 230 |
-
|
| 231 |
-
|
| 232 |
-
|
| 233 |
-
|
| 234 |
-
|
| 235 |
-
|
|
|
|
| 236 |
|
| 237 |
-
else:
|
| 238 |
-
|
| 239 |
-
|
| 240 |
-
|
| 241 |
-
|
| 242 |
-
|
| 243 |
-
|
| 244 |
-
|
| 245 |
-
|
| 246 |
-
|
| 247 |
-
|
| 248 |
-
|
| 249 |
-
|
| 250 |
-
|
| 251 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 252 |
|
| 253 |
-
if threshold <= 0.90:
|
| 254 |
-
context_list = sentence_id_combine(data, query_results, lag=window)
|
| 255 |
else:
|
| 256 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 257 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 258 |
|
| 259 |
-
if decoder_model == "GPT3 - (text-davinci-003)":
|
| 260 |
-
prompt = generate_gpt_prompt(query_text, context_list)
|
| 261 |
with col2:
|
| 262 |
with st.form("my_form"):
|
| 263 |
edited_prompt = st.text_area(
|
|
@@ -273,9 +389,20 @@ if decoder_model == "GPT3 - (text-davinci-003)":
|
|
| 273 |
if submitted:
|
| 274 |
api_key = save_key(openai_key)
|
| 275 |
openai.api_key = api_key
|
| 276 |
-
generated_text =
|
| 277 |
st.subheader("Answer:")
|
| 278 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 279 |
|
| 280 |
|
| 281 |
elif decoder_model == "T5":
|
|
@@ -384,22 +511,53 @@ if decoder_model == "GPT-J":
|
|
| 384 |
)
|
| 385 |
submitted = st.form_submit_button("Submit")
|
| 386 |
|
|
|
|
| 387 |
|
| 388 |
-
with col1:
|
| 389 |
-
with st.expander("See Retrieved Text"):
|
| 390 |
-
st.subheader("Retrieved Text:")
|
| 391 |
-
for context_text in context_list:
|
| 392 |
-
context_text = f"""{context_text}"""
|
| 393 |
-
st.write(
|
| 394 |
-
f"<ul><li><p>{context_text}</p></li></ul>",
|
| 395 |
-
unsafe_allow_html=True,
|
| 396 |
-
)
|
| 397 |
-
|
| 398 |
-
file_text = retrieve_transcript(data, year, quarter, ticker)
|
| 399 |
|
| 400 |
-
with
|
| 401 |
-
|
| 402 |
-
st.
|
| 403 |
-
|
| 404 |
-
|
| 405 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import re
|
| 2 |
+
|
| 3 |
import openai
|
| 4 |
import streamlit_scrollable_textbox as stx
|
| 5 |
|
|
|
|
| 27 |
get_spacy_model,
|
| 28 |
get_splade_sparse_embedding_model,
|
| 29 |
get_t5_model,
|
| 30 |
+
gpt_turbo_model,
|
| 31 |
save_key,
|
| 32 |
)
|
| 33 |
from utils.prompts import (
|
|
|
|
| 38 |
generate_flant5_prompt_summ_chunk_context_single,
|
| 39 |
generate_gpt_j_two_shot_prompt_1,
|
| 40 |
generate_gpt_j_two_shot_prompt_2,
|
| 41 |
+
generate_gpt_prompt_alpaca,
|
| 42 |
+
generate_gpt_prompt_alpaca_multi_doc,
|
| 43 |
+
generate_gpt_prompt_original,
|
| 44 |
+
generate_multi_doc_context,
|
| 45 |
get_context_list_prompt,
|
| 46 |
)
|
| 47 |
from utils.retriever import (
|
|
|
|
| 50 |
query_pinecone_sparse,
|
| 51 |
sentence_id_combine,
|
| 52 |
text_lookup,
|
| 53 |
+
year_quarter_range,
|
| 54 |
)
|
| 55 |
from utils.transcript_retrieval import retrieve_transcript
|
| 56 |
from utils.vector_index import (
|
|
|
|
| 71 |
|
| 72 |
with st.sidebar:
|
| 73 |
ner_choice = st.selectbox("Select NER Model", ["Spacy", "Alpaca"])
|
| 74 |
+
document_type = st.selectbox(
|
| 75 |
+
"Select Query Type", ["Single-Document", "Multi-Document"]
|
| 76 |
+
)
|
| 77 |
|
| 78 |
if ner_choice == "Spacy":
|
| 79 |
ner_model = get_spacy_model()
|
| 80 |
|
| 81 |
with col1:
|
| 82 |
st.subheader("Question")
|
| 83 |
+
if document_type == "Single-Document":
|
| 84 |
+
query_text = st.text_area(
|
| 85 |
+
"Input Query",
|
| 86 |
+
value="What was discussed regarding Wearables revenue performance?",
|
| 87 |
+
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 88 |
else:
|
| 89 |
+
query_text = st.text_area(
|
| 90 |
+
"Input Query",
|
| 91 |
+
value="How has Apple's revenue from Wearables performed over the past 2 years?",
|
| 92 |
)
|
| 93 |
|
|
|
|
|
|
|
| 94 |
|
| 95 |
+
years_choice = ["2020", "2019", "2018", "2017", "2016", "All"]
|
| 96 |
+
quarters_choice = ["Q1", "Q2", "Q3", "Q4", "All"]
|
| 97 |
ticker_choice = [
|
| 98 |
"AAPL",
|
| 99 |
"CSCO",
|
|
|
|
| 107 |
"AMD",
|
| 108 |
]
|
| 109 |
|
| 110 |
+
|
| 111 |
+
if document_type == "Single-Document":
|
| 112 |
+
if ner_choice == "Alpaca":
|
| 113 |
+
ner_prompt = generate_alpaca_ner_prompt(query_text)
|
| 114 |
+
entity_text = generate_entities_flan_alpaca_inference_api(ner_prompt)
|
| 115 |
+
company_ent, quarter_ent, year_ent = format_entities_flan_alpaca(
|
| 116 |
+
entity_text
|
| 117 |
+
)
|
| 118 |
else:
|
| 119 |
+
company_ent = extract_ticker_spacy(query_text, ner_model)
|
| 120 |
+
quarter_ent, year_ent = extract_quarter_year(query_text)
|
| 121 |
+
|
| 122 |
+
ticker_index, quarter_index, year_index = clean_entities(
|
| 123 |
+
company_ent, quarter_ent, year_ent
|
| 124 |
+
)
|
| 125 |
+
|
| 126 |
+
with col1:
|
| 127 |
+
# Hardcoding the defaults for a question without metadata
|
| 128 |
+
if (
|
| 129 |
+
query_text
|
| 130 |
+
== "What was discussed regarding Wearables revenue performance?"
|
| 131 |
+
):
|
| 132 |
+
year = st.selectbox("Year", years_choice)
|
| 133 |
+
quarter = st.selectbox("Quarter", quarters_choice)
|
| 134 |
+
ticker = st.selectbox("Company", ticker_choice)
|
| 135 |
+
else:
|
| 136 |
+
year = st.selectbox("Year", years_choice, index=year_index)
|
| 137 |
+
quarter = st.selectbox(
|
| 138 |
+
"Quarter", quarters_choice, index=quarter_index
|
| 139 |
+
)
|
| 140 |
+
ticker = st.selectbox("Company", ticker_choice, ticker_index)
|
| 141 |
+
|
| 142 |
+
participant_type = st.selectbox(
|
| 143 |
+
"Speaker", ["Company Speaker", "Analyst"]
|
| 144 |
+
)
|
| 145 |
+
|
| 146 |
+
else:
|
| 147 |
+
# Multi-Document Case
|
| 148 |
+
|
| 149 |
+
with col1:
|
| 150 |
+
# Hardcoding the defaults for a question without metadata
|
| 151 |
+
if (
|
| 152 |
+
query_text
|
| 153 |
+
== "How has Apple's revenue from Wearables performed over the past 2 years?"
|
| 154 |
+
):
|
| 155 |
+
start_year = st.selectbox("Start Year", years_choice, index=2)
|
| 156 |
+
start_quarter = st.selectbox(
|
| 157 |
+
"Start Quarter", quarters_choice, index=0
|
| 158 |
+
)
|
| 159 |
+
|
| 160 |
+
end_year = st.selectbox("End Year", years_choice, index=0)
|
| 161 |
+
end_quarter = st.selectbox("End Quarter", quarters_choice, index=0)
|
| 162 |
+
|
| 163 |
+
ticker = st.selectbox("Company", ticker_choice, index=0)
|
| 164 |
+
else:
|
| 165 |
+
start_year = st.selectbox("Start Year", years_choice, index=2)
|
| 166 |
+
start_quarter = st.selectbox(
|
| 167 |
+
"Start Quarter", quarters_choice, index=0
|
| 168 |
+
)
|
| 169 |
+
|
| 170 |
+
end_year = st.selectbox("End Year", years_choice, index=0)
|
| 171 |
+
end_quarter = st.selectbox("End Quarter", quarters_choice, index=0)
|
| 172 |
+
|
| 173 |
+
ticker = st.selectbox("Company", ticker_choice, index=0)
|
| 174 |
+
|
| 175 |
+
participant_type = st.selectbox(
|
| 176 |
+
"Speaker", ["Company Speaker", "Analyst"]
|
| 177 |
+
)
|
| 178 |
+
|
| 179 |
|
| 180 |
with st.sidebar:
|
| 181 |
st.subheader("Select Options:")
|
| 182 |
|
| 183 |
+
if document_type == "Single-Document":
|
| 184 |
+
num_results = int(
|
| 185 |
+
st.number_input("Number of Results to query", 1, 15, value=5)
|
| 186 |
+
)
|
| 187 |
+
else:
|
| 188 |
+
num_results = int(
|
| 189 |
+
st.number_input("Number of Results to query", 1, 15, value=2)
|
| 190 |
+
)
|
| 191 |
|
| 192 |
|
| 193 |
# Choose encoder model
|
|
|
|
| 199 |
|
| 200 |
# Choose decoder model
|
| 201 |
|
| 202 |
+
# Restricting multi-document to only GPT-3
|
| 203 |
+
if document_type == "Single-Document":
|
| 204 |
+
decoder_models_choice = ["GPT-3.5 Turbo", "T5", "FLAN-T5", "GPT-J"]
|
| 205 |
+
else:
|
| 206 |
+
decoder_models_choice = ["GPT-3.5 Turbo"]
|
| 207 |
with st.sidebar:
|
| 208 |
decoder_model = st.selectbox("Select Decoder Model", decoder_models_choice)
|
| 209 |
|
|
|
|
| 240 |
) = get_splade_sparse_embedding_model()
|
| 241 |
|
| 242 |
with st.sidebar:
|
| 243 |
+
if document_type == "Single-Document":
|
| 244 |
+
window = int(st.number_input("Sentence Window Size", 0, 10, value=1))
|
| 245 |
+
|
| 246 |
+
threshold = float(
|
| 247 |
+
st.number_input(
|
| 248 |
+
label="Similarity Score Threshold",
|
| 249 |
+
step=0.05,
|
| 250 |
+
format="%.2f",
|
| 251 |
+
value=0.25,
|
| 252 |
+
)
|
| 253 |
+
)
|
| 254 |
+
else:
|
| 255 |
+
window = int(st.number_input("Sentence Window Size", 0, 10, value=0))
|
| 256 |
+
|
| 257 |
+
threshold = float(
|
| 258 |
+
st.number_input(
|
| 259 |
+
label="Similarity Score Threshold",
|
| 260 |
+
step=0.05,
|
| 261 |
+
format="%.2f",
|
| 262 |
+
value=0.6,
|
| 263 |
+
)
|
| 264 |
)
|
|
|
|
| 265 |
|
| 266 |
data = get_data()
|
| 267 |
|
| 268 |
+
if document_type == "Single-Document":
|
| 269 |
+
if encoder_model == "Hybrid SGPT - SPLADE":
|
| 270 |
+
dense_query_embedding = create_dense_embeddings(
|
| 271 |
+
query_text, retriever_model
|
| 272 |
+
)
|
| 273 |
+
sparse_query_embedding = create_sparse_embeddings(
|
| 274 |
+
query_text, sparse_retriever_model, sparse_retriever_tokenizer
|
| 275 |
+
)
|
| 276 |
+
dense_query_embedding, sparse_query_embedding = hybrid_score_norm(
|
| 277 |
+
dense_query_embedding, sparse_query_embedding, 0
|
| 278 |
+
)
|
| 279 |
+
query_results = query_pinecone_sparse(
|
| 280 |
+
dense_query_embedding,
|
| 281 |
+
sparse_query_embedding,
|
| 282 |
+
num_results,
|
| 283 |
+
pinecone_index,
|
| 284 |
+
year,
|
| 285 |
+
quarter,
|
| 286 |
+
ticker,
|
| 287 |
+
participant_type,
|
| 288 |
+
threshold,
|
| 289 |
+
)
|
| 290 |
|
| 291 |
+
else:
|
| 292 |
+
dense_query_embedding = create_dense_embeddings(
|
| 293 |
+
query_text, retriever_model
|
| 294 |
+
)
|
| 295 |
+
query_results = query_pinecone(
|
| 296 |
+
dense_query_embedding,
|
| 297 |
+
num_results,
|
| 298 |
+
pinecone_index,
|
| 299 |
+
year,
|
| 300 |
+
quarter,
|
| 301 |
+
ticker,
|
| 302 |
+
participant_type,
|
| 303 |
+
threshold,
|
| 304 |
+
)
|
| 305 |
|
| 306 |
+
if threshold <= 0.90:
|
| 307 |
+
context_list = sentence_id_combine(data, query_results, lag=window)
|
| 308 |
+
else:
|
| 309 |
+
context_list = format_query(query_results)
|
| 310 |
|
|
|
|
|
|
|
| 311 |
else:
|
| 312 |
+
# Multi-Document Retreival
|
| 313 |
+
if encoder_model == "Hybrid SGPT - SPLADE":
|
| 314 |
+
dense_query_embedding = create_dense_embeddings(
|
| 315 |
+
query_text, retriever_model
|
| 316 |
+
)
|
| 317 |
+
sparse_query_embedding = create_sparse_embeddings(
|
| 318 |
+
query_text, sparse_retriever_model, sparse_retriever_tokenizer
|
| 319 |
+
)
|
| 320 |
+
dense_query_embedding, sparse_query_embedding = hybrid_score_norm(
|
| 321 |
+
dense_query_embedding, sparse_query_embedding, 0
|
| 322 |
+
)
|
| 323 |
+
year_quarter_list = year_quarter_range(
|
| 324 |
+
start_quarter, start_year, end_quarter, end_year
|
| 325 |
+
)
|
| 326 |
+
|
| 327 |
+
context_group = []
|
| 328 |
+
for year, quarter in year_quarter_list:
|
| 329 |
+
query_results = query_pinecone_sparse(
|
| 330 |
+
dense_query_embedding,
|
| 331 |
+
sparse_query_embedding,
|
| 332 |
+
num_results,
|
| 333 |
+
pinecone_index,
|
| 334 |
+
year,
|
| 335 |
+
quarter,
|
| 336 |
+
ticker,
|
| 337 |
+
participant_type,
|
| 338 |
+
threshold,
|
| 339 |
+
)
|
| 340 |
+
results_list = sentence_id_combine(data, query_results, lag=window)
|
| 341 |
+
context_group.append((results_list, year, quarter))
|
| 342 |
+
|
| 343 |
+
else:
|
| 344 |
+
dense_query_embedding = create_dense_embeddings(
|
| 345 |
+
query_text, retriever_model
|
| 346 |
+
)
|
| 347 |
+
year_quarter_list = year_quarter_range(
|
| 348 |
+
start_quarter, start_year, end_quarter, end_year
|
| 349 |
+
)
|
| 350 |
+
|
| 351 |
+
context_group = []
|
| 352 |
+
for year, quarter in year_quarter_list:
|
| 353 |
+
query_results = query_pinecone(
|
| 354 |
+
dense_query_embedding,
|
| 355 |
+
num_results,
|
| 356 |
+
pinecone_index,
|
| 357 |
+
year,
|
| 358 |
+
quarter,
|
| 359 |
+
ticker,
|
| 360 |
+
participant_type,
|
| 361 |
+
threshold,
|
| 362 |
+
)
|
| 363 |
+
results_list = sentence_id_combine(data, query_results, lag=window)
|
| 364 |
+
context_group.append((results_list, year, quarter))
|
| 365 |
|
| 366 |
+
multi_doc_context = generate_multi_doc_context(context_group)
|
| 367 |
+
|
| 368 |
+
|
| 369 |
+
if decoder_model == "GPT-3.5 Turbo":
|
| 370 |
+
if document_type == "Single-Document":
|
| 371 |
+
prompt = generate_gpt_prompt_alpaca(query_text, context_list)
|
| 372 |
+
else:
|
| 373 |
+
prompt = generate_gpt_prompt_alpaca_multi_doc(
|
| 374 |
+
query_text, context_group
|
| 375 |
+
)
|
| 376 |
|
|
|
|
|
|
|
| 377 |
with col2:
|
| 378 |
with st.form("my_form"):
|
| 379 |
edited_prompt = st.text_area(
|
|
|
|
| 389 |
if submitted:
|
| 390 |
api_key = save_key(openai_key)
|
| 391 |
openai.api_key = api_key
|
| 392 |
+
generated_text = gpt_turbo_model(edited_prompt)
|
| 393 |
st.subheader("Answer:")
|
| 394 |
+
regex_pattern_sentences = (
|
| 395 |
+
"(?<!\w\.\w.)(?<![A-Z][a-z]\.)(?<=\.|\?)\s"
|
| 396 |
+
)
|
| 397 |
+
generated_text_list = re.split(
|
| 398 |
+
regex_pattern_sentences, generated_text
|
| 399 |
+
)
|
| 400 |
+
for answer_text in generated_text_list:
|
| 401 |
+
answer_text = f"""{answer_text}"""
|
| 402 |
+
st.write(
|
| 403 |
+
f"<ul><li><p>{answer_text}</p></li></ul>",
|
| 404 |
+
unsafe_allow_html=True,
|
| 405 |
+
)
|
| 406 |
|
| 407 |
|
| 408 |
elif decoder_model == "T5":
|
|
|
|
| 511 |
)
|
| 512 |
submitted = st.form_submit_button("Submit")
|
| 513 |
|
| 514 |
+
tab1, tab2 = st.tabs(["Retrived Text", "Retrieved Documents"])
|
| 515 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 516 |
|
| 517 |
+
with tab1:
|
| 518 |
+
if document_type == "Single-Document":
|
| 519 |
+
with st.expander("See Retrieved Text"):
|
| 520 |
+
st.subheader("Retrieved Text:")
|
| 521 |
+
for context_text in context_list:
|
| 522 |
+
context_text = f"""{context_text}"""
|
| 523 |
+
st.write(
|
| 524 |
+
f"<ul><li><p>{context_text}</p></li></ul>",
|
| 525 |
+
unsafe_allow_html=True,
|
| 526 |
+
)
|
| 527 |
+
else:
|
| 528 |
+
with st.expander("See Retrieved Text"):
|
| 529 |
+
st.subheader("Retrieved Text:")
|
| 530 |
+
sections = [
|
| 531 |
+
s.strip()
|
| 532 |
+
for s in multi_doc_context.split("Document: ")
|
| 533 |
+
if s.strip()
|
| 534 |
+
]
|
| 535 |
+
|
| 536 |
+
# Add "Document: " back to the beginning of each section
|
| 537 |
+
context_list = [
|
| 538 |
+
"Document: " + s[0:7] + "\n" + s[7:] for s in sections
|
| 539 |
+
]
|
| 540 |
+
for context_text in context_list:
|
| 541 |
+
context_text = f"""{context_text}"""
|
| 542 |
+
st.write(
|
| 543 |
+
f"<ul><li><p>{context_text}</p></li></ul>",
|
| 544 |
+
unsafe_allow_html=True,
|
| 545 |
+
)
|
| 546 |
+
|
| 547 |
+
|
| 548 |
+
with tab2:
|
| 549 |
+
if document_type == "Single-Document":
|
| 550 |
+
file_text = retrieve_transcript(data, year, quarter, ticker)
|
| 551 |
+
with st.expander("See Transcript"):
|
| 552 |
+
st.subheader("Earnings Call Transcript:")
|
| 553 |
+
stx.scrollableTextbox(
|
| 554 |
+
file_text, height=700, border=False, fontFamily="Helvetica"
|
| 555 |
+
)
|
| 556 |
+
else:
|
| 557 |
+
for year, quarter in year_quarter_list:
|
| 558 |
+
file_text = retrieve_transcript(data, year, quarter, ticker)
|
| 559 |
+
with st.expander(f"See Transcript - {quarter} {year}"):
|
| 560 |
+
st.subheader("Earnings Call Transcript - {quarter} {year}:")
|
| 561 |
+
stx.scrollableTextbox(
|
| 562 |
+
file_text, height=700, border=False, fontFamily="Helvetica"
|
| 563 |
+
)
|
utils/models.py
CHANGED
|
@@ -103,14 +103,16 @@ def save_key(api_key):
|
|
| 103 |
# Text Generation
|
| 104 |
|
| 105 |
|
| 106 |
-
def
|
| 107 |
-
response = openai.
|
| 108 |
-
model="
|
| 109 |
-
|
| 110 |
-
|
|
|
|
|
|
|
| 111 |
max_tokens=1024,
|
| 112 |
)
|
| 113 |
-
return response
|
| 114 |
|
| 115 |
|
| 116 |
def generate_text_flan_t5(model, tokenizer, input_text):
|
|
|
|
| 103 |
# Text Generation
|
| 104 |
|
| 105 |
|
| 106 |
+
def gpt_turbo_model(prompt):
|
| 107 |
+
response = openai.ChatCompletion.create(
|
| 108 |
+
model="gpt-3.5-turbo",
|
| 109 |
+
messages=[
|
| 110 |
+
{"role": "user", "content": prompt},
|
| 111 |
+
],
|
| 112 |
+
temperature=0.01,
|
| 113 |
max_tokens=1024,
|
| 114 |
)
|
| 115 |
+
return response["choices"][0]["message"]["content"]
|
| 116 |
|
| 117 |
|
| 118 |
def generate_text_flan_t5(model, tokenizer, input_text):
|
utils/prompts.py
CHANGED
|
@@ -1,4 +1,51 @@
|
|
| 1 |
-
def
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2 |
context = " ".join(context_list)
|
| 3 |
prompt = f"""Answer the question in 6 long detailed points as accurately as possible using the provided context. Include as many key details as possible.
|
| 4 |
Context: {context}
|
|
|
|
| 1 |
+
def generate_multi_doc_context(context_group):
|
| 2 |
+
multi_doc_context = ""
|
| 3 |
+
for context_text_list, year, quarter in context_group:
|
| 4 |
+
print((context_text_list, year, quarter))
|
| 5 |
+
if context_text_list == []:
|
| 6 |
+
break
|
| 7 |
+
else:
|
| 8 |
+
multi_doc_context = (
|
| 9 |
+
multi_doc_context
|
| 10 |
+
+ "\n"
|
| 11 |
+
+ f"Document: {quarter} {year}"
|
| 12 |
+
+ "\n"
|
| 13 |
+
+ " ".join(context_text_list)
|
| 14 |
+
)
|
| 15 |
+
return multi_doc_context
|
| 16 |
+
|
| 17 |
+
|
| 18 |
+
def generate_gpt_prompt_alpaca(query_text, context_list):
|
| 19 |
+
context = " ".join(context_list)
|
| 20 |
+
prompt = f"""Below is an instruction that describes a task, paired with an input that provides further context. Use the following guidelines to write a response that that appropriately completes the request:
|
| 21 |
+
### Instruction:
|
| 22 |
+
- Write a detailed paragraph consisting of exactly five complete sentences that answer the question based on the provided context.
|
| 23 |
+
- Focus on addressing the specific question posed, providing as much relevant information and detail as possible.
|
| 24 |
+
- Only use details from the provided context that directly address the question; do not include any additional information that is not explicitly stated.
|
| 25 |
+
- Aim to provide a clear and concise summary that fully addresses the question.
|
| 26 |
+
|
| 27 |
+
Question: {query_text}
|
| 28 |
+
Context: {context}
|
| 29 |
+
### Response:"""
|
| 30 |
+
return prompt
|
| 31 |
+
|
| 32 |
+
|
| 33 |
+
def generate_gpt_prompt_alpaca_multi_doc(query_text, context_group):
|
| 34 |
+
multi_doc_context = generate_multi_doc_context(context_group)
|
| 35 |
+
prompt = f"""Below is an instruction that describes a task, paired with an input that provides further context. Use the following guidelines to write a response that that appropriately completes the request:
|
| 36 |
+
### Instruction:
|
| 37 |
+
- Write a detailed paragraph consisting of exactly five complete sentences that answer the question based on the provided context.
|
| 38 |
+
- Focus on addressing the specific question posed, providing as much relevant information and detail as possible.
|
| 39 |
+
- Only use details from the provided context that directly address the question; do not include any additional information that is not explicitly stated.
|
| 40 |
+
- Aim to provide a clear and concise summary that fully addresses the question.
|
| 41 |
+
|
| 42 |
+
Question: {query_text}
|
| 43 |
+
Context: {multi_doc_context}
|
| 44 |
+
### Response:"""
|
| 45 |
+
return prompt
|
| 46 |
+
|
| 47 |
+
|
| 48 |
+
def generate_gpt_prompt_original(query_text, context_list):
|
| 49 |
context = " ".join(context_list)
|
| 50 |
prompt = f"""Answer the question in 6 long detailed points as accurately as possible using the provided context. Include as many key details as possible.
|
| 51 |
Context: {context}
|
utils/retriever.py
CHANGED
|
@@ -195,3 +195,55 @@ def sentence_id_combine(data, query_results, lag=1):
|
|
| 195 |
def text_lookup(data, sentence_ids):
|
| 196 |
context = ". ".join(data.iloc[sentence_ids].to_list())
|
| 197 |
return context
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 195 |
def text_lookup(data, sentence_ids):
|
| 196 |
context = ". ".join(data.iloc[sentence_ids].to_list())
|
| 197 |
return context
|
| 198 |
+
|
| 199 |
+
|
| 200 |
+
def year_quarter_range(start_quarter, start_year, end_quarter, end_year):
|
| 201 |
+
"""Creates a list of all (year, quarter) pairs that lie in the range including the start and end quarters."""
|
| 202 |
+
start_year = int(start_year)
|
| 203 |
+
end_year = int(end_year)
|
| 204 |
+
|
| 205 |
+
quarters = (
|
| 206 |
+
[("Q1", "Q2", "Q3", "Q4")] * (end_year - start_year)
|
| 207 |
+
+ [("Q1", "Q2", "Q3" if end_quarter == "Q4" else "Q4")]
|
| 208 |
+
* (end_quarter == "Q4")
|
| 209 |
+
+ [
|
| 210 |
+
(
|
| 211 |
+
"Q1"
|
| 212 |
+
if start_quarter == "Q1"
|
| 213 |
+
else "Q2"
|
| 214 |
+
if start_quarter == "Q2"
|
| 215 |
+
else "Q3"
|
| 216 |
+
if start_quarter == "Q3"
|
| 217 |
+
else "Q4",
|
| 218 |
+
)
|
| 219 |
+
* (end_year - start_year)
|
| 220 |
+
]
|
| 221 |
+
)
|
| 222 |
+
years = list(range(start_year, end_year + 1))
|
| 223 |
+
list_year_quarter = [
|
| 224 |
+
(y, q) for y in years for q in quarters[years.index(y)]
|
| 225 |
+
]
|
| 226 |
+
# Remove duplicate pairs
|
| 227 |
+
seen = set()
|
| 228 |
+
list_year_quarter_cleaned = []
|
| 229 |
+
for tup in list_year_quarter:
|
| 230 |
+
if tup not in seen:
|
| 231 |
+
seen.add(tup)
|
| 232 |
+
list_year_quarter_cleaned.append(tup)
|
| 233 |
+
return list_year_quarter_cleaned
|
| 234 |
+
|
| 235 |
+
|
| 236 |
+
def multi_document_query(
|
| 237 |
+
dense_query_embedding,
|
| 238 |
+
sparse_query_embedding,
|
| 239 |
+
num_results,
|
| 240 |
+
pinecone_index,
|
| 241 |
+
start_quarter,
|
| 242 |
+
start_year,
|
| 243 |
+
end_quarter,
|
| 244 |
+
end_year,
|
| 245 |
+
ticker,
|
| 246 |
+
participant_type,
|
| 247 |
+
threshold,
|
| 248 |
+
):
|
| 249 |
+
pass
|