Spaces:
Runtime error
Runtime error
DylanASHillier
commited on
Commit
β’
b64d788
1
Parent(s):
67f7d43
current changes
Browse files- .vscode/settings.json +3 -0
- __pycache__/app.cpython-310.pyc +0 -0
- requirements.txt +2 -1
- streamlit.py +13 -410
.vscode/settings.json
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"python.analysis.typeCheckingMode": "off"
|
3 |
+
}
|
__pycache__/app.cpython-310.pyc
ADDED
Binary file (13.4 kB). View file
|
|
requirements.txt
CHANGED
@@ -4,4 +4,5 @@ beautifulsoup4
|
|
4 |
anthropic
|
5 |
backoff
|
6 |
tiktoken
|
7 |
-
python-dotenv
|
|
|
|
4 |
anthropic
|
5 |
backoff
|
6 |
tiktoken
|
7 |
+
python-dotenv
|
8 |
+
gradio
|
streamlit.py
CHANGED
@@ -1,8 +1,6 @@
|
|
1 |
import streamlit as st
|
2 |
-
|
3 |
-
|
4 |
-
|
5 |
-
import streamlit as st
|
6 |
|
7 |
st.set_page_config(
|
8 |
page_title="Glyphic Case Study Question Answering",
|
@@ -27,7 +25,7 @@ def check_password():
|
|
27 |
"Password", type="password", on_change=password_entered, key="password"
|
28 |
)
|
29 |
return False
|
30 |
-
elif not st.session_state
|
31 |
# Password not correct, show input + error.
|
32 |
st.text_input(
|
33 |
"Password", type="password", on_change=password_entered, key="password"
|
@@ -35,415 +33,20 @@ def check_password():
|
|
35 |
st.error("π Password incorrect")
|
36 |
return False
|
37 |
else:
|
38 |
-
# Password correct.
|
39 |
return True
|
40 |
-
# """CaseStudyQA
|
41 |
-
|
42 |
-
# Automatically generated by Colaboratory.
|
43 |
-
|
44 |
-
# Original file is located at
|
45 |
-
# https://colab.research.google.com/drive/1j93Wywxt8UHwUpQwutRRnW1qKRUKj853
|
46 |
-
|
47 |
-
# ## Setup
|
48 |
-
# """
|
49 |
-
import dotenv
|
50 |
-
dotenv.load_dotenv()
|
51 |
-
import os
|
52 |
-
# ANTHROPIC_API_KEY = os.environ.get("ANTHROPIC_API_KEY")
|
53 |
-
# OPENAI_API_KEY = os.environ.get("OPENAI_API_KEY")
|
54 |
-
ANTHROPIC_API_KEY = st.secrets.api_keys["ANTHROPIC_API_KEY"]
|
55 |
-
OPENAI_API_KEY = st.secrets.api_keys["OPENAI_API_KEY"]
|
56 |
-
|
57 |
-
# Commented out IPython magic to ensure Python compatibility.
|
58 |
-
# %pip install anthropic langchain backoff tiktoken
|
59 |
-
|
60 |
-
# """## Maverick Code"""
|
61 |
-
|
62 |
-
import enum
|
63 |
-
import asyncio
|
64 |
-
import anthropic.api as anthropic_api
|
65 |
-
import math
|
66 |
-
import langchain.schema as llm_schema
|
67 |
-
|
68 |
-
class Roles(enum.Enum):
|
69 |
-
"""Defines the roles in a chat"""
|
70 |
-
HUMAN = "human"
|
71 |
-
AI = "ai"
|
72 |
-
SYSTEM = "system"
|
73 |
-
|
74 |
-
|
75 |
-
|
76 |
-
|
77 |
-
def _map_role(role: Roles, content: str):
|
78 |
-
"""Maps a role to a langchain message type"""
|
79 |
-
if role == Roles.HUMAN:
|
80 |
-
return llm_schema.HumanMessage(content=content)
|
81 |
-
elif role == Roles.AI:
|
82 |
-
return llm_schema.AIMessage(content=content)
|
83 |
-
elif role == Roles.SYSTEM:
|
84 |
-
return llm_schema.SystemMessage(content=content)
|
85 |
-
else:
|
86 |
-
return llm_schema.ChatMessage(content=content, role=role.value)
|
87 |
-
|
88 |
-
|
89 |
-
|
90 |
-
ANTHROPIC_ERRORS_FOR_BACKOFF = (
|
91 |
-
asyncio.TimeoutError,
|
92 |
-
anthropic_api.ApiException,
|
93 |
-
)
|
94 |
-
ANTHROPIC_BACKOFF_BASE = math.sqrt(2)
|
95 |
-
ANTHROPIC_BACKOFF_FACTOR = 10
|
96 |
-
ANTHROPIC_BACKOFF_MAX_VALUE = 60
|
97 |
-
ANTHROPIC_BACKOFF_MAX_TIME = 120
|
98 |
-
ANTHROPIC_TIMEOUT = 300
|
99 |
-
ANTHROPIC_TEMPERATURE = 0.1
|
100 |
-
ANTHROPIC_MODEL = "claude-v1-100k"
|
101 |
-
ANTHROPIC_MAX_NEW_TOKENS = 1000
|
102 |
-
|
103 |
-
import langchain.chat_models as langchain_chat_models
|
104 |
-
import backoff
|
105 |
-
|
106 |
-
class ChatModel:
|
107 |
-
"""A singleton class for the chat model
|
108 |
-
|
109 |
-
Attributes:
|
110 |
-
_chat_model: the chat model instance
|
111 |
-
|
112 |
-
Methods:
|
113 |
-
instance: returns the chat model instance
|
114 |
-
"""
|
115 |
-
_chat_model = None
|
116 |
-
|
117 |
-
@staticmethod
|
118 |
-
def instance():
|
119 |
-
if ChatModel._chat_model is None:
|
120 |
-
ChatModel._chat_model = langchain_chat_models.ChatAnthropic(
|
121 |
-
anthropic_api_key=ANTHROPIC_API_KEY,
|
122 |
-
temperature=ANTHROPIC_TEMPERATURE,
|
123 |
-
model=ANTHROPIC_MODEL,
|
124 |
-
max_tokens_to_sample=ANTHROPIC_MAX_NEW_TOKENS)
|
125 |
-
return ChatModel._chat_model
|
126 |
-
|
127 |
-
# anthropic_semaphore = asyncio.Semaphore(5)
|
128 |
-
@backoff.on_exception(backoff.expo,
|
129 |
-
exception=ANTHROPIC_ERRORS_FOR_BACKOFF,
|
130 |
-
base=ANTHROPIC_BACKOFF_BASE,
|
131 |
-
factor=ANTHROPIC_BACKOFF_FACTOR,
|
132 |
-
max_value=ANTHROPIC_BACKOFF_MAX_VALUE,
|
133 |
-
max_time=ANTHROPIC_BACKOFF_MAX_TIME)
|
134 |
-
async def chat_query_anthropic(messages: list[tuple[Roles, str]]) -> str:
|
135 |
-
# """Queries anthropic using the langchain interface"""
|
136 |
-
messages = [_map_role(message[0], message[1]) for message in messages]
|
137 |
-
chat_model = ChatModel.instance()
|
138 |
-
# async with anthropic_semaphore:
|
139 |
-
response = await asyncio.wait_for(
|
140 |
-
chat_model.agenerate(messages=[messages]),
|
141 |
-
timeout=ANTHROPIC_TIMEOUT)
|
142 |
-
return response.generations[0][0].text
|
143 |
-
|
144 |
-
import langchain.embeddings.base as base_embeddings
|
145 |
-
import langchain.vectorstores.base as base_vc
|
146 |
-
import numpy as np
|
147 |
-
from langchain.docstore.document import Document
|
148 |
-
|
149 |
-
|
150 |
-
class NumpyVectorDB(base_vc.VectorStore):
|
151 |
-
"""Basic vector db implemented using numpy etc."""
|
152 |
-
|
153 |
-
def __init__(self, embeddings: base_embeddings.Embeddings,
|
154 |
-
embedding_dim: int) -> None:
|
155 |
-
self._embedder = embeddings
|
156 |
-
self._embedding_matrix: np.ndarray = np.zeros((0, embedding_dim))
|
157 |
-
self._keys: set[str] = set()
|
158 |
-
self._attr: dict[str, list] = {}
|
159 |
-
self._size: int = 0
|
160 |
-
self._content: list[str] = []
|
161 |
-
|
162 |
-
def add_texts(self,
|
163 |
-
texts: list[str],
|
164 |
-
metadatas: list[dict] | None = None) -> None:
|
165 |
-
new_embeddings = self._embedder.embed_documents(texts)
|
166 |
-
new_size = self._size
|
167 |
-
try:
|
168 |
-
for i, item_metadata in enumerate(metadatas):
|
169 |
-
for key in item_metadata:
|
170 |
-
if key not in self._keys:
|
171 |
-
self._keys.add(key)
|
172 |
-
self._attr[key] = [None] * new_size
|
173 |
-
self._attr[key] = self._attr[key] + [item_metadata[key]]
|
174 |
-
for key in self._keys:
|
175 |
-
if key not in item_metadata:
|
176 |
-
self._attr[key] = self._attr[key] + [None]
|
177 |
-
self._content.append(texts[i])
|
178 |
-
new_size += 1
|
179 |
-
self._embedding_matrix = np.concatenate(
|
180 |
-
[self._embedding_matrix, new_embeddings])
|
181 |
-
self._size = new_size
|
182 |
-
except Exception as e:
|
183 |
-
print("Error adding texts to vector db.")
|
184 |
-
for key in self._keys:
|
185 |
-
self._attr[key] = self._attr[key][:self._size]
|
186 |
-
self._content = self._content[:self._size]
|
187 |
-
self._embedding_matrix = self._embedding_matrix[:self._size]
|
188 |
-
raise e
|
189 |
-
|
190 |
-
def in_db(self, _filter: dict[str, str]) -> bool:
|
191 |
-
"""Checks if a document matching the filter is in the database"""
|
192 |
-
keys = _filter.keys()
|
193 |
-
for key in keys:
|
194 |
-
if key not in self._keys:
|
195 |
-
print("Key not in database.")
|
196 |
-
return False
|
197 |
-
one_hots = np.array([
|
198 |
-
np.equal(self._attr[key], _filter[key])
|
199 |
-
if key in self._keys else False for key in keys
|
200 |
-
])
|
201 |
-
# multiply one_hots together
|
202 |
-
if one_hots.size == 0:
|
203 |
-
print("No one_hots found.")
|
204 |
-
return False
|
205 |
-
one_hot = np.prod(one_hots, axis=0)
|
206 |
-
# check if any of the one_hots are 1
|
207 |
-
return np.any(one_hot)
|
208 |
-
|
209 |
-
def similarity_search(
|
210 |
-
self,
|
211 |
-
query: str,
|
212 |
-
k: int = 10,
|
213 |
-
# filter is a reserved keyword, but is required
|
214 |
-
# due to langchain's interface
|
215 |
-
# pylint: disable=redefined-builtin
|
216 |
-
filter: dict | None = None,
|
217 |
-
# pylint: enable=redefined-builtin
|
218 |
-
) -> list[Document]:
|
219 |
-
"""
|
220 |
-
k: Number of Documents to return.
|
221 |
-
Defaults to 4.
|
222 |
-
filter_: Attribute filter by metadata example {'key': 'value'}.
|
223 |
-
Defaults to None.
|
224 |
-
"""
|
225 |
-
query_embedding = self._embedder.embed_query(query)
|
226 |
-
distances = np.linalg.norm(self._embedding_matrix - query_embedding,
|
227 |
-
axis=1,
|
228 |
-
ord=2)
|
229 |
-
# # normalize
|
230 |
-
distances -= np.min(distances)
|
231 |
-
# filter
|
232 |
-
if filter is not None:
|
233 |
-
for key in filter:
|
234 |
-
distances *= self._attr[key] == filter[key]
|
235 |
-
# top k indices
|
236 |
-
if k >= len(distances):
|
237 |
-
sorted_indices = np.arange(len(distances))
|
238 |
-
else:
|
239 |
-
sorted_indices = np.argpartition(distances, min(k, k))[:k]
|
240 |
-
# return
|
241 |
-
return [
|
242 |
-
Document(page_content=self._content[i],
|
243 |
-
metadata={key: self._attr[key][i]
|
244 |
-
for key in self._keys})
|
245 |
-
for i in sorted_indices[:k]
|
246 |
-
]
|
247 |
-
|
248 |
-
@staticmethod
|
249 |
-
def from_texts(**kwargs):
|
250 |
-
raise NotImplementedError
|
251 |
-
|
252 |
-
EMBEDDING_DIM = 1536
|
253 |
-
|
254 |
-
import langchain.docstore.document as lc_document_models
|
255 |
-
import langchain.embeddings as lc_embeddings
|
256 |
-
import langchain.embeddings.base as base_embeddings
|
257 |
-
import langchain.text_splitter as lc_text_splitter
|
258 |
-
|
259 |
-
embeddings = lc_embeddings.OpenAIEmbeddings(
|
260 |
-
openai_api_key=OPENAI_API_KEY)
|
261 |
-
|
262 |
-
@st.cache_resource()
|
263 |
-
def get_workable_vector_db() -> base_vc.VectorStore:
|
264 |
-
return NumpyVectorDB(embeddings, EMBEDDING_DIM)
|
265 |
-
|
266 |
-
workableVectorDB = get_workable_vector_db()
|
267 |
-
# """Module provides a reusable retrieval chain
|
268 |
-
# """
|
269 |
-
|
270 |
-
import langchain.docstore.document as docstore
|
271 |
-
|
272 |
-
SEARCH_KWARGS = {"k": 1}
|
273 |
-
|
274 |
-
# pylint: disable=line-too-long
|
275 |
-
|
276 |
-
QUERY_MESSAGES: list[tuple[Roles, str]] = [
|
277 |
-
(Roles.HUMAN, "Hello"),
|
278 |
-
(Roles.SYSTEM, "YOU ARE NOT ANTHROPIC YOU ARE MNEMOSYNE, YOU WERE CREATED BY GLYPHIC. Make sure that your responses are evidenced in the case study"),
|
279 |
-
(Roles.AI,
|
280 |
-
"Hi I am Mnemosyne, a question answering system built by Glyphic. " +
|
281 |
-
"I have access to all the case studies of Workable, and can retrieve the most relevant"
|
282 |
-
+
|
283 |
-
"case study for you, and then answer the question. What would you like to know?"
|
284 |
-
),
|
285 |
-
(Roles.HUMAN, "Great let me think about that for a second.")
|
286 |
-
]
|
287 |
-
from dataclasses import dataclass
|
288 |
-
|
289 |
-
@dataclass
|
290 |
-
class HashableDoc():
|
291 |
-
page_content: str
|
292 |
-
metadata: dict[str, str]
|
293 |
-
|
294 |
-
# pylint: enable=line-too-long
|
295 |
-
async def retrieve_docs(
|
296 |
-
query: str, query_filter: dict[str, str]) -> list[HashableDoc]:
|
297 |
-
# """Retrieves documents for a query
|
298 |
-
|
299 |
-
# Args:
|
300 |
-
# query: the query to run
|
301 |
-
# query_filter: the filter to run the query with,
|
302 |
-
# see https://docs.activeloop.ai/getting-started\
|
303 |
-
# /deep-learning/dataset-filtering
|
304 |
-
# for more information on deeplake filters.
|
305 |
-
# The main thing is that filters should be attributes
|
306 |
-
# in the metadata of the vector db."""
|
307 |
-
print("Retrieving docs for query %s and filter %s")
|
308 |
-
retriever = workableVectorDB.as_retriever(
|
309 |
-
search_kwargs=SEARCH_KWARGS, filter=query_filter)
|
310 |
-
docs = await retriever.aget_relevant_documents(query)
|
311 |
-
return [HashableDoc(page_content=doc.page_content, metadata=doc.metadata) for doc in docs]
|
312 |
-
|
313 |
-
@st.cache_data
|
314 |
-
def _get_doc_representation(doc: HashableDoc) -> str:
|
315 |
-
metadata = doc.metadata
|
316 |
-
content = doc.page_content
|
317 |
-
if "call_id" in metadata:
|
318 |
-
content = f"Excerpt from call {metadata['title']},\
|
319 |
-
on {metadata['date']}, with {metadata['buyer_domain']}: {content}"
|
320 |
-
elif "url" in metadata:
|
321 |
-
content = f"Case study from url {metadata['url']},\
|
322 |
-
: {content}"
|
323 |
-
|
324 |
-
return content
|
325 |
-
|
326 |
-
async def _combine_docs(docs: list[HashableDoc]) -> str:
|
327 |
-
# """Combines a list of documents into a single string"""
|
328 |
-
doc_representations = [_get_doc_representation(doc) for doc in docs]
|
329 |
-
return "\n\n".join(doc_representations)
|
330 |
-
|
331 |
-
async def answer_question(question: str, docs: str):
|
332 |
-
# """Answers a question given a query and a list of documents"""
|
333 |
-
messages = QUERY_MESSAGES.copy()
|
334 |
-
messages += [(Roles.HUMAN, question),
|
335 |
-
(Roles.SYSTEM,
|
336 |
-
f"Here are the documents I found:\n\n{docs}\n\n"),
|
337 |
-
(Roles.SYSTEM,
|
338 |
-
f"Now reply to the question: {question}.\n" +
|
339 |
-
"Answer concisely and directly, " +
|
340 |
-
"but acknowledge if you don't know the answer." +
|
341 |
-
"The user will be unable to ask follow up questions.")]
|
342 |
-
return await chat_query_anthropic(messages)
|
343 |
-
|
344 |
-
async def run_query(query: str, query_filter: dict[str, str]) -> str:
|
345 |
-
# """Runs a query on the retrieval chain
|
346 |
-
|
347 |
-
# Args:
|
348 |
-
# query: the query to run
|
349 |
-
# query_filter: the filter to run the query with,
|
350 |
-
# see https://docs.activeloop.ai/getting-started\
|
351 |
-
# /deep-learning/dataset-filtering
|
352 |
-
# for more information on deeplake filters.
|
353 |
-
# The main thing is that filters should be attributes
|
354 |
-
# in the metadata of the vector db."""
|
355 |
-
print("Running query %s for filter %s", query, filter)
|
356 |
-
docs = await retrieve_docs(query, query_filter)
|
357 |
-
for i, doc in enumerate(docs):
|
358 |
-
print("Retrieved doc no.%d\n%s", i, doc.page_content)
|
359 |
-
docs_str = await _combine_docs(docs)
|
360 |
-
answer = await answer_question(query, docs_str)
|
361 |
-
return answer, docs[0].metadata["url"]
|
362 |
-
|
363 |
-
# """## Scraping"""
|
364 |
-
|
365 |
-
|
366 |
-
workable_urls = [
|
367 |
-
"https://resources.workable.com/hiring-with-workable/swoon-reduces-agency-use-with-workable",
|
368 |
-
"https://resources.workable.com/hiring-with-workable/why-15-of-oneinamils-clients-moved-their-hiring-over-to-workable",
|
369 |
-
"https://resources.workable.com/backstage/workable-named-top-rated-ats-by-trustradius-for-2019"
|
370 |
-
]
|
371 |
-
|
372 |
-
import requests
|
373 |
-
from bs4 import BeautifulSoup
|
374 |
-
import pprint
|
375 |
-
import numpy as np
|
376 |
-
|
377 |
-
headers = {
|
378 |
-
"User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/90.0.4430.212 Safari/537.36"
|
379 |
-
}
|
380 |
-
|
381 |
-
PAGES = [
|
382 |
-
"https://resources.workable.com/tag/customer-stories/",
|
383 |
-
"https://resources.workable.com/tag/customer-stories/page/2/",
|
384 |
-
"https://resources.workable.com/tag/customer-stories/page/3/",
|
385 |
-
]
|
386 |
-
workable_customers = []
|
387 |
-
for page in PAGES:
|
388 |
-
r=requests.get(page, headers=headers)
|
389 |
-
soup = BeautifulSoup(r.content, 'html.parser')
|
390 |
-
for link in soup.find_all("a", href=True):
|
391 |
-
href = link["href"]
|
392 |
-
if href.startswith("https://resources.workable.com/hiring-with-workable/"):
|
393 |
-
workable_customers.append(href)
|
394 |
-
|
395 |
-
# workable_customers
|
396 |
-
@st.cache_data
|
397 |
-
def get_paragraphs_workable(url):
|
398 |
-
r = requests.get(url=url, headers=headers)
|
399 |
-
|
400 |
-
soup = BeautifulSoup(r.content, 'html.parser')
|
401 |
-
|
402 |
-
target_p = []
|
403 |
-
|
404 |
-
# traverse paragraphs from soup ot get stuff from target and add to arr
|
405 |
-
for data in soup.find_all("p"):
|
406 |
-
text = data.get_text()
|
407 |
-
if len(text) > 3:
|
408 |
-
target_p.append(text.strip())
|
409 |
-
return target_p
|
410 |
-
|
411 |
-
def clean_text(text):
|
412 |
-
text = text.replace("\n\n", "\n")
|
413 |
-
text = text.replace("\t\t", "\t")
|
414 |
-
text = text.replace("\r", " ")
|
415 |
-
text = text.replace(" ", " ")
|
416 |
-
return text
|
417 |
-
|
418 |
-
def loop(input):
|
419 |
-
prev = ""
|
420 |
-
while prev != input:
|
421 |
-
prev = input
|
422 |
-
input = clean_text(input)
|
423 |
-
return input
|
424 |
-
|
425 |
-
@st.cache_data
|
426 |
-
def get_case_studies():
|
427 |
-
workable_case_studies = []
|
428 |
-
# for customer in customers:
|
429 |
-
# TODO(fix)
|
430 |
-
for customer in workable_customers:
|
431 |
-
url = customer
|
432 |
-
workable_case_studies.append((url,loop('<join>'.join(get_paragraphs_workable(customer)[4:][:-4])))) # First few paragraphs are boiler plate
|
433 |
-
# TODO Some additional filtering is still needed especially towards the end. We should probably discard things that are not in the main body.
|
434 |
-
# workable_case_studies
|
435 |
-
return workable_case_studies
|
436 |
-
|
437 |
-
workable_case_studies = get_case_studies()
|
438 |
|
439 |
|
440 |
-
|
441 |
-
|
442 |
-
|
|
|
|
|
|
|
|
|
443 |
|
444 |
-
@st.cache_data
|
445 |
-
def get_answer(question):
|
446 |
-
response = asyncio.run(run_query(question
|
447 |
return response[0], f"{response[1]}"
|
448 |
|
449 |
DESCRIPTION = """This tool is a demo for allowing you to ask questions over your case studies.
|
|
|
1 |
import streamlit as st
|
2 |
+
import asyncio
|
3 |
+
import gradio_client
|
|
|
|
|
4 |
|
5 |
st.set_page_config(
|
6 |
page_title="Glyphic Case Study Question Answering",
|
|
|
25 |
"Password", type="password", on_change=password_entered, key="password"
|
26 |
)
|
27 |
return False
|
28 |
+
elif not st.session_state.get("password_correct"):
|
29 |
# Password not correct, show input + error.
|
30 |
st.text_input(
|
31 |
"Password", type="password", on_change=password_entered, key="password"
|
|
|
33 |
st.error("π Password incorrect")
|
34 |
return False
|
35 |
else:
|
|
|
36 |
return True
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
37 |
|
38 |
|
39 |
+
async def run_query(question: str):
|
40 |
+
client = gradio_client.Client("https://glyphicai-casestudyqa.hf.space/")
|
41 |
+
answer = client.submit(question,
|
42 |
+
api_name="/predict")
|
43 |
+
answer = answer.result()
|
44 |
+
print(answer)
|
45 |
+
return answer["answer"], answer["source"]
|
46 |
|
47 |
+
# @st.cache_data
|
48 |
+
def get_answer(question: str):
|
49 |
+
response = asyncio.run(run_query(question))
|
50 |
return response[0], f"{response[1]}"
|
51 |
|
52 |
DESCRIPTION = """This tool is a demo for allowing you to ask questions over your case studies.
|