File size: 12,798 Bytes
e375940
 
 
94a67ea
8cd1f1e
94a67ea
 
 
c5e4524
9c49e99
6a79fd2
 
c5e4524
5482130
c5e4524
 
244a3e0
 
c5e4524
 
 
 
 
 
 
 
 
 
 
6627aee
 
0175cb6
6627aee
0175cb6
6627aee
 
0175cb6
c5e4524
0175cb6
9c49e99
c5e4524
 
 
f9da573
e375940
f9da573
 
 
c5e4524
 
 
 
 
 
8cd1f1e
ff5dcc7
a7b0635
9975133
b19bb41
 
 
8cd1f1e
9975133
 
9c49e99
0175cb6
0f10e25
0175cb6
244a3e0
0175cb6
9c49e99
9975133
 
6627aee
637ca1b
6627aee
637ca1b
8cd1f1e
0175cb6
5482130
244a3e0
6a79fd2
 
 
0175cb6
6a79fd2
 
0175cb6
9c49e99
 
 
 
9975133
ac5b87a
c5f41e6
9975133
6627aee
 
 
 
 
 
 
 
c5f41e6
9975133
6627aee
 
 
 
 
 
 
 
 
 
ac5b87a
 
 
c5f41e6
 
 
 
 
 
 
 
 
 
 
 
 
 
9975133
6627aee
 
 
 
 
 
 
 
c5f41e6
9975133
 
 
 
e375940
94a67ea
e375940
8cd1f1e
 
 
 
e375940
9975133
 
8cd1f1e
 
 
 
6627aee
8cd1f1e
9975133
 
8cd1f1e
 
 
 
e375940
 
 
8cd1f1e
 
 
 
 
 
e375940
 
 
8cd1f1e
 
 
 
e375940
 
 
 
 
 
 
 
 
 
 
 
8cd1f1e
9975133
bf8b612
fbd690d
9975133
 
 
e375940
 
 
 
9975133
e514fa8
 
fbd690d
 
e375940
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e514fa8
872808d
e514fa8
 
 
8cd1f1e
 
2a99161
9c49e99
9975133
 
e375940
 
 
9975133
 
 
 
 
 
 
 
 
 
 
9c49e99
 
 
8cd1f1e
 
0175cb6
 
 
8cd1f1e
 
9975133
9c49e99
 
 
 
 
 
 
 
 
 
 
 
 
 
8cd1f1e
 
6627aee
8cd1f1e
9975133
6627aee
0175cb6
 
6627aee
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9c49e99
 
 
 
 
 
6627aee
0175cb6
 
 
6627aee
 
 
 
0175cb6
 
 
6627aee
0175cb6
 
 
 
6627aee
 
 
 
 
 
0175cb6
 
 
 
 
6627aee
0175cb6
 
 
 
6627aee
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9c49e99
8cd1f1e
9975133
 
ff5dcc7
9975133
ff5dcc7
 
 
 
 
f9da573
 
 
9975133
 
ff5dcc7
9975133
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
import openai
import streamlit_scrollable_textbox as stx

import pinecone
import streamlit as st

st.set_page_config(layout="wide")  # isort: split

from utils.entity_extraction import (
    clean_entities,
    extract_quarter_year,
    extract_ticker_spacy,
    format_entities_flan_alpaca,
    generate_alpaca_ner_prompt,
)
from utils.models import (
    generate_entities_flan_alpaca_checkpoint,
    generate_entities_flan_alpaca_inference_api,
    generate_text_flan_t5,
    get_data,
    get_flan_alpaca_xl_model,
    get_flan_t5_model,
    get_mpnet_embedding_model,
    get_sgpt_embedding_model,
    get_spacy_model,
    get_splade_sparse_embedding_model,
    get_t5_model,
)
from utils.prompts import (
    generate_flant5_prompt_instruct_chunk_context,
    generate_flant5_prompt_instruct_chunk_context_single,
    generate_flant5_prompt_instruct_complete_context,
    generate_flant5_prompt_summ_chunk_context,
    generate_flant5_prompt_summ_chunk_context_single,
    generate_gpt_j_two_shot_prompt_1,
    generate_gpt_j_two_shot_prompt_2,
    generate_gpt_prompt,
    generate_gpt_prompt_2,
    generate_text_flan_t5,
    get_context_list_prompt,
)
from utils.retriever import (
    format_query,
    query_pinecone,
    query_pinecone_sparse,
    sentence_id_combine,
    text_lookup,
)
from utils.transcript_retrieval import retrieve_transcript
from utils.vector_index import (
    create_dense_embeddings,
    create_sparse_embeddings,
    hybrid_score_norm,
)

st.title("Question Answering on Earnings Call Transcripts")


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."
)

col1, col2 = st.columns([3, 3], gap="medium")


with st.sidebar:
    ner_choice = st.selectbox("Select NER Model", ["Spacy", "Alpaca"])

if ner_choice == "Spacy":
    ner_model = get_spacy_model()

with col1:
    st.subheader("Question")
    query_text = st.text_area(
        "Input Query",
        value="What was discussed regarding Wearables revenue performance?",
    )

if ner_choice == "Alpaca":
    ner_prompt = generate_alpaca_ner_prompt(query_text)
    entity_text = generate_entities_flan_alpaca_inference_api(ner_prompt)
    company_ent, quarter_ent, year_ent = format_entities_flan_alpaca(
        entity_text
    )
else:
    company_ent = extract_ticker_spacy(query_text, ner_model)
    quarter_ent, year_ent = extract_quarter_year(query_text)

ticker_index, quarter_index, year_index = clean_entities(
    company_ent, quarter_ent, year_ent
)

with col1:
    years_choice = ["2020", "2019", "2018", "2017", "2016", "All"]

with col1:
    # Hardcoding the defaults for a question without metadata
    if (
        query_text
        == "What was discussed regarding Wearables revenue performance?"
    ):
        year = st.selectbox("Year", years_choice)
    else:
        year = st.selectbox("Year", years_choice, index=year_index)

with col1:
    # Hardcoding the defaults for a question without metadata
    if (
        query_text
        == "What was discussed regarding Wearables revenue performance?"
    ):
        quarter = st.selectbox("Quarter", ["Q1", "Q2", "Q3", "Q4", "All"])
    else:
        quarter = st.selectbox(
            "Quarter", ["Q1", "Q2", "Q3", "Q4", "All"], index=quarter_index
        )

with col1:
    participant_type = st.selectbox("Speaker", ["Company Speaker", "Analyst"])

ticker_choice = [
    "AAPL",
    "CSCO",
    "MSFT",
    "ASML",
    "NVDA",
    "GOOGL",
    "MU",
    "INTC",
    "AMZN",
    "AMD",
]

with col1:
    # Hardcoding the defaults for a question without metadata
    if (
        query_text
        == "What was discussed regarding Wearables revenue performance?"
    ):
        ticker = st.selectbox("Company", ticker_choice)
    else:
        ticker = st.selectbox("Company", ticker_choice, ticker_index)

with st.sidebar:
    st.subheader("Select Options:")

with st.sidebar:
    num_results = int(
        st.number_input("Number of Results to query", 1, 15, value=5)
    )


# Choose encoder model

encoder_models_choice = ["MPNET", "SGPT", "Hybrid MPNET - SPLADE"]
with st.sidebar:
    encoder_model = st.selectbox("Select Encoder Model", encoder_models_choice)


# Choose decoder model

decoder_models_choice = ["GPT3 - (text-davinci-003)", "T5", "FLAN-T5", "GPT-J"]

with st.sidebar:
    decoder_model = st.selectbox("Select Decoder Model", decoder_models_choice)


if encoder_model == "MPNET":
    # Connect to pinecone environment
    pinecone.init(
        api_key=st.secrets["pinecone_mpnet"], environment="us-east1-gcp"
    )
    pinecone_index_name = "week2-all-mpnet-base"
    pinecone_index = pinecone.Index(pinecone_index_name)
    retriever_model = get_mpnet_embedding_model()

elif encoder_model == "SGPT":
    # Connect to pinecone environment
    pinecone.init(
        api_key=st.secrets["pinecone_sgpt"], environment="us-east1-gcp"
    )
    pinecone_index_name = "week2-sgpt-125m"
    pinecone_index = pinecone.Index(pinecone_index_name)
    retriever_model = get_sgpt_embedding_model()

elif encoder_model == "Hybrid MPNET - SPLADE":
    pinecone.init(
        api_key=st.secrets["pinecone_hybrid_splade_mpnet"],
        environment="us-central1-gcp",
    )
    pinecone_index_name = "splade-mpnet"
    pinecone_index = pinecone.Index(pinecone_index_name)
    retriever_model = get_mpnet_embedding_model()
    (
        sparse_retriever_model,
        sparse_retriever_tokenizer,
    ) = get_splade_sparse_embedding_model()

with st.sidebar:
    window = int(st.number_input("Sentence Window Size", 0, 10, value=1))

with st.sidebar:
    threshold = float(
        st.number_input(
            label="Similarity Score Threshold",
            step=0.05,
            format="%.2f",
            value=0.25,
        )
    )

data = get_data()

if encoder_model == "Hybrid SGPT - SPLADE":
    dense_query_embedding = create_dense_embeddings(
        query_text, retriever_model
    )
    sparse_query_embedding = create_sparse_embeddings(
        query_text, sparse_retriever_model, sparse_retriever_tokenizer
    )
    dense_query_embedding, sparse_query_embedding = hybrid_score_norm(
        dense_query_embedding, sparse_query_embedding, 0
    )
    query_results = query_pinecone_sparse(
        dense_query_embedding,
        sparse_query_embedding,
        num_results,
        pinecone_index,
        year,
        quarter,
        ticker,
        participant_type,
        threshold,
    )

else:
    dense_query_embedding = create_dense_embeddings(
        query_text, retriever_model
    )
    query_results = query_pinecone(
        dense_query_embedding,
        num_results,
        pinecone_index,
        year,
        quarter,
        ticker,
        participant_type,
        threshold,
    )


if threshold <= 0.90:
    context_list = sentence_id_combine(data, query_results, lag=window)
else:
    context_list = format_query(query_results)


if decoder_model == "GPT3 - (text-davinci-003)":
    prompt = generate_gpt_prompt(query_text, context_list)
    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)
                st.subheader("Answer:")
                st.write(generated_text)


elif decoder_model == "T5":
    prompt = generate_flant5_prompt_instruct_complete_context(
        query_text, context_list
    )
    t5_pipeline = get_t5_model()
    output_text = []
    with col2:
        with st.form("my_form"):
            edited_prompt = st.text_area(
                label="Model Prompt", value=prompt, height=270
            )
            context_list = get_context_list_prompt(edited_prompt)
            submitted = st.form_submit_button("Submit")
            if submitted:
                for context_text in context_list:
                    output_text.append(
                        t5_pipeline(context_text)[0]["summary_text"]
                    )
                st.subheader("Answer:")
                for text in output_text:
                    st.markdown(f"- {text}")

elif decoder_model == "FLAN-T5":
    flan_t5_model, flan_t5_tokenizer = get_flan_t5_model()
    output_text = []
    with col2:
        prompt_type = st.selectbox(
            "Select prompt type",
            ["Complete Text QA", "Chunkwise QA", "Chunkwise Summarize"],
        )
        if prompt_type == "Complete Text QA":
            prompt = generate_flant5_prompt_instruct_complete_context(
                query_text, context_list
            )
        elif prompt_type == "Chunkwise QA":
            st.write("The following prompt is not editable.")
            prompt = generate_flant5_prompt_instruct_chunk_context(
                query_text, context_list
            )
        elif prompt_type == "Chunkwise Summarize":
            st.write("The following prompt is not editable.")
            prompt = generate_flant5_prompt_summ_chunk_context(
                query_text, context_list
            )
        else:
            prompt = ""
        with st.form("my_form"):
            edited_prompt = st.text_area(
                label="Model Prompt", value=prompt, height=270
            )
            submitted = st.form_submit_button("Submit")
            if submitted:
                if prompt_type == "Complete Text QA":
                    output_text_string = generate_text_flan_t5(
                        flan_t5_model, flan_t5_tokenizer, prompt
                    )
                    st.subheader("Answer:")
                    st.write(output_text_string)
                elif prompt_type == "Chunkwise QA":
                    for context_text in context_list:
                        model_input = generate_flant5_prompt_instruct_chunk_context_single(
                            query_text, context_text
                        )
                        output_text.append(
                            generate_text_flan_t5(
                                flan_t5_model, flan_t5_tokenizer, model_input
                            )
                        )
                    st.subheader("Answer:")
                    for text in output_text:
                        if "(iii)" not in text:
                            st.markdown(f"- {text}")
                elif prompt_type == "Chunkwise Summarize":
                    for context_text in context_list:
                        model_input = (
                            generate_flant5_prompt_summ_chunk_context_single(
                                query_text, context_text
                            )
                        )
                        output_text.append(
                            generate_text_flan_t5(
                                flan_t5_model, flan_t5_tokenizer, model_input
                            )
                        )
                    st.subheader("Answer:")
                    for text in output_text:
                        if "(iii)" not in text:
                            st.markdown(f"- {text}")

if decoder_model == "GPT-J":
    if ticker in ["AAPL", "AMD"]:
        prompt = generate_gpt_j_two_shot_prompt_1(query_text, context_list)
    elif ticker in ["NVDA", "INTC", "AMZN"]:
        prompt = generate_gpt_j_two_shot_prompt_2(query_text, context_list)
    else:
        prompt = generate_gpt_j_two_shot_prompt_1(query_text, context_list)
    with col2:
        with st.form("my_form"):
            edited_prompt = st.text_area(
                label="Model Prompt", value=prompt, height=270
            )
            st.write(
                "The app currently just shows the prompt. The app does not load the model due to memory limitations."
            )
            submitted = st.form_submit_button("Submit")


with col1:
    with st.expander("See Retrieved Text"):
        st.subheader("Retrieved Text:")
        for context_text in context_list:
            context_text = f"""{context_text}"""
            st.write(
                f"<ul><li><p>{context_text}</p></li></ul>",
                unsafe_allow_html=True,
            )

file_text = retrieve_transcript(data, year, quarter, ticker)

with col1:
    with st.expander("See Transcript"):
        st.subheader("Earnings Call Transcript:")
        stx.scrollableTextbox(
            file_text, height=700, border=False, fontFamily="Helvetica"
        )