Spaces:
Build error
Build error
File size: 11,131 Bytes
e375940 f9da573 e375940 f9da573 e375940 f9da573 e375940 f9da573 e375940 f9da573 e375940 f9da573 8d46199 f9da573 e375940 f9da573 e375940 f9da573 e375940 ac5b87a e375940 ac5b87a 6966109 ac5b87a 6966109 ac5b87a e375940 ac5b87a e375940 ac5b87a e375940 ac5b87a f9da573 e375940 f9da573 ac5b87a f9da573 e375940 f9da573 e375940 f9da573 e375940 f9da573 9975133 bf8b612 9b8f482 bf8b612 2b221a0 bf8b612 9975133 8d46199 9975133 f9da573 9975133 f9da573 bf8b612 f9da573 bf8b612 f9da573 6966109 f9da573 2b221a0 f9da573 8d46199 f9da573 |
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 |
import openai
import pandas as pd
import streamlit_scrollable_textbox as stx
import torch
from sentence_transformers import SentenceTransformer
from tqdm import tqdm
from transformers import (
AutoModelForMaskedLM,
AutoModelForSeq2SeqLM,
AutoTokenizer,
pipeline,
)
import pinecone
import streamlit as st
@st.experimental_singleton
def get_data():
data = pd.read_csv("earnings_calls_cleaned_metadata.csv")
return data
# Initialize models from HuggingFace
@st.experimental_singleton
def get_t5_model():
return pipeline("summarization", model="t5-small", tokenizer="t5-small")
@st.experimental_singleton
def get_flan_t5_model():
return pipeline(
"summarization",
model="google/flan-t5-small",
tokenizer="google/flan-t5-small",
max_length=512,
# length_penalty = 0
)
@st.experimental_singleton
def get_mpnet_embedding_model():
device = "cuda" if torch.cuda.is_available() else "cpu"
model = SentenceTransformer(
"sentence-transformers/all-mpnet-base-v2", device=device
)
model.max_seq_length = 512
return model
@st.experimental_singleton
def get_splade_sparse_embedding_model():
model_sparse = "naver/splade-cocondenser-ensembledistil"
# check device
device = "cuda" if torch.cuda.is_available() else "cpu"
tokenizer = AutoTokenizer.from_pretrained(model_sparse)
model_sparse = AutoModelForMaskedLM.from_pretrained(model_sparse)
# move to gpu if available
model_sparse.to(device)
return model_sparse, tokenizer
@st.experimental_singleton
def get_sgpt_embedding_model():
device = "cuda" if torch.cuda.is_available() else "cpu"
model = SentenceTransformer(
"Muennighoff/SGPT-125M-weightedmean-nli-bitfit", device=device
)
model.max_seq_length = 512
return model
@st.experimental_memo
def save_key(api_key):
return api_key
def create_dense_embeddings(query, model):
dense_emb = model.encode([query]).tolist()
return dense_emb
def create_sparse_embeddings(query, model, tokenizer):
device = "cuda" if torch.cuda.is_available() else "cpu"
inputs = tokenizer(query, return_tensors="pt").to(device)
with torch.no_grad():
logits = model(**inputs).logits
inter = torch.log1p(torch.relu(logits[0]))
token_max = torch.max(inter, dim=0) # sum over input tokens
nz_tokens = torch.where(token_max.values > 0)[0]
nz_weights = token_max.values[nz_tokens]
order = torch.sort(nz_weights, descending=True)
nz_weights = nz_weights[order[1]]
nz_tokens = nz_tokens[order[1]]
return {
"indices": nz_tokens.cpu().numpy().tolist(),
"values": nz_weights.cpu().numpy().tolist(),
}
def hybrid_score_norm(dense, sparse, alpha: float):
"""Hybrid score using a convex combination
alpha * dense + (1 - alpha) * sparse
Args:
dense: Array of floats representing
sparse: a dict of `indices` and `values`
alpha: scale between 0 and 1
"""
if alpha < 0 or alpha > 1:
raise ValueError("Alpha must be between 0 and 1")
hs = {
"indices": sparse["indices"],
"values": [v * (1 - alpha) for v in sparse["values"]],
}
return [v * alpha for v in dense], hs
def query_pinecone_sparse(
dense_vec,
sparse_vec,
top_k,
index,
year,
quarter,
ticker,
participant_type,
threshold=0.25,
):
if participant_type == "Company Speaker":
participant = "Answer"
else:
participant = "Question"
if year == "All":
if quarter == "All":
xc = index.query(
vector=dense_vec,
sparse_vector=sparse_vec,
top_k=top_k,
filter={
"Year": {
"$in": [
int("2020"),
int("2019"),
int("2018"),
int("2017"),
int("2016"),
]
},
"Quarter": {"$in": ["Q1", "Q2", "Q3", "Q4"]},
"Ticker": {"$eq": ticker},
"QA_Flag": {"$eq": participant},
},
include_metadata=True,
)
else:
xc = index.query(
vector=dense_vec,
sparse_vector=sparse_vec,
top_k=top_k,
filter={
"Year": {
"$in": [
int("2020"),
int("2019"),
int("2018"),
int("2017"),
int("2016"),
]
},
"Quarter": {"$eq": quarter},
"Ticker": {"$eq": ticker},
"QA_Flag": {"$eq": participant},
},
include_metadata=True,
)
else:
# search pinecone index for context passage with the answer
xc = index.query(
vector=dense_vec,
sparse_vector=sparse_vec,
top_k=top_k,
filter={
"Year": int(year),
"Quarter": {"$eq": quarter},
"Ticker": {"$eq": ticker},
"QA_Flag": {"$eq": participant},
},
include_metadata=True,
)
# filter the context passages based on the score threshold
filtered_matches = []
for match in xc["matches"]:
if match["score"] >= threshold:
filtered_matches.append(match)
xc["matches"] = filtered_matches
return xc
def query_pinecone(
dense_vec,
top_k,
index,
year,
quarter,
ticker,
participant_type,
threshold=0.25,
):
if participant_type == "Company Speaker":
participant = "Answer"
else:
participant = "Question"
if year == "All":
if quarter == "All":
xc = index.query(
vector=dense_vec,
top_k=top_k,
filter={
"Year": {
"$in": [
int("2020"),
int("2019"),
int("2018"),
int("2017"),
int("2016"),
]
},
"Quarter": {"$in": ["Q1", "Q2", "Q3", "Q4"]},
"Ticker": {"$eq": ticker},
"QA_Flag": {"$eq": participant},
},
include_metadata=True,
)
else:
xc = index.query(
vector=dense_vec,
top_k=top_k,
filter={
"Year": {
"$in": [
int("2020"),
int("2019"),
int("2018"),
int("2017"),
int("2016"),
]
},
"Quarter": {"$eq": quarter},
"Ticker": {"$eq": ticker},
"QA_Flag": {"$eq": participant},
},
include_metadata=True,
)
else:
# search pinecone index for context passage with the answer
xc = index.query(
vector=dense_vec,
top_k=top_k,
filter={
"Year": int(year),
"Quarter": {"$eq": quarter},
"Ticker": {"$eq": ticker},
"QA_Flag": {"$eq": participant},
},
include_metadata=True,
)
# filter the context passages based on the score threshold
filtered_matches = []
for match in xc["matches"]:
if match["score"] >= threshold:
filtered_matches.append(match)
xc["matches"] = filtered_matches
return xc
def format_query(query_results):
# extract passage_text from Pinecone search result
context = [
result["metadata"]["Text"] for result in query_results["matches"]
]
return context
def sentence_id_combine(data, query_results, lag=1):
# Extract sentence IDs from query results
ids = [
result["metadata"]["Sentence_id"]
for result in query_results["matches"]
]
# Generate new IDs by adding a lag value to the original IDs
new_ids = [id + i for id in ids for i in range(-lag, lag + 1)]
# Remove duplicates and sort the new IDs
new_ids = sorted(set(new_ids))
# Create a list of lookup IDs by grouping the new IDs in groups of lag*2+1
lookup_ids = [
new_ids[i : i + (lag * 2 + 1)]
for i in range(0, len(new_ids), lag * 2 + 1)
]
# Create a list of context sentences by joining the sentences corresponding to the lookup IDs
context_list = [
" ".join(
data.loc[data["Sentence_id"].isin(lookup_id), "Text"].to_list()
)
for lookup_id in lookup_ids
]
return context_list
def text_lookup(data, sentence_ids):
context = ". ".join(data.iloc[sentence_ids].to_list())
return context
def generate_prompt(query_text, context_list):
context = " ".join(context_list)
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.
Context: {context}
Question: {query_text}
Answer:"""
return prompt
def generate_prompt_2(query_text, context_list):
context = " ".join(context_list)
prompt = f"""
Context information is below:
---------------------
{context}
---------------------
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=prompt,
temperature=0.1,
max_tokens=1024,
top_p=1.0,
frequency_penalty=0.5,
presence_penalty=1,
)
return response.choices[0].text
# Transcript Retrieval
def retrieve_transcript(data, year, quarter, ticker):
if year == "All" or quarter == "All":
row = (
data.loc[
(data.Ticker == ticker),
["File_Name"],
]
.drop_duplicates()
.iloc[0, 0]
)
else:
row = (
data.loc[
(data.Year == int(year))
& (data.Quarter == quarter)
& (data.Ticker == ticker),
["File_Name"],
]
.drop_duplicates()
.iloc[0, 0]
)
# convert row to a string and join values with "-"
# row_str = "-".join(row.astype(str)) + ".txt"
open_file = open(
f"Transcripts/{ticker}/{row}",
"r",
)
file_text = open_file.read()
return file_text
|