Spaces:
Running
on
CPU Upgrade
Running
on
CPU Upgrade
Update app.py
Browse files
app.py
CHANGED
@@ -8,9 +8,16 @@ import re
|
|
8 |
import json
|
9 |
from auditqa.sample_questions import QUESTIONS
|
10 |
from auditqa.reports import POSSIBLE_REPORTS
|
11 |
-
from auditqa.engine.prompts import audience_prompts
|
12 |
from auditqa.doc_process import process_pdf
|
13 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
14 |
|
15 |
async def chat(query,history,audience,sources,reports):
|
16 |
"""taking a query and a message history, use a pipeline (reformulation, retriever, answering) to yield a tuple of:
|
@@ -21,6 +28,9 @@ async def chat(query,history,audience,sources,reports):
|
|
21 |
print(f"audience:{audience}")
|
22 |
print(f"sources:{sources}")
|
23 |
print(f"reports:{reports}")
|
|
|
|
|
|
|
24 |
|
25 |
if audience == "Children":
|
26 |
audience_prompt = audience_prompts["children"]
|
@@ -33,20 +43,101 @@ async def chat(query,history,audience,sources,reports):
|
|
33 |
|
34 |
# Prepare default values
|
35 |
if len(sources) == 0:
|
36 |
-
sources = ["
|
37 |
|
38 |
if len(reports) == 0:
|
39 |
reports = []
|
40 |
|
41 |
-
|
|
|
|
|
|
|
42 |
|
43 |
-
|
44 |
-
|
45 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
46 |
|
47 |
yield history,docs_html,output_query,output_language
|
48 |
|
49 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
50 |
# --------------------------------------------------------------------
|
51 |
# Gradio
|
52 |
# --------------------------------------------------------------------
|
|
|
8 |
import json
|
9 |
from auditqa.sample_questions import QUESTIONS
|
10 |
from auditqa.reports import POSSIBLE_REPORTS
|
11 |
+
from auditqa.engine.prompts import audience_prompts, answer_prompt_template
|
12 |
from auditqa.doc_process import process_pdf
|
13 |
+
from langchain_core.prompts import ChatPromptTemplate
|
14 |
+
from langchain_core.output_parsers import StrOutputParser
|
15 |
+
from langchain.llms import HuggingFaceEndpoint
|
16 |
+
from dotenv import load_dotenv
|
17 |
+
load_dotenv()
|
18 |
+
|
19 |
+
HF_token = os.environ["HF_TOKEN"]
|
20 |
+
vectorstores = process_pdf()
|
21 |
|
22 |
async def chat(query,history,audience,sources,reports):
|
23 |
"""taking a query and a message history, use a pipeline (reformulation, retriever, answering) to yield a tuple of:
|
|
|
28 |
print(f"audience:{audience}")
|
29 |
print(f"sources:{sources}")
|
30 |
print(f"reports:{reports}")
|
31 |
+
docs_html = ""
|
32 |
+
output_query = ""
|
33 |
+
output_language = "english"
|
34 |
|
35 |
if audience == "Children":
|
36 |
audience_prompt = audience_prompts["children"]
|
|
|
43 |
|
44 |
# Prepare default values
|
45 |
if len(sources) == 0:
|
46 |
+
sources = ["ABC"]
|
47 |
|
48 |
if len(reports) == 0:
|
49 |
reports = []
|
50 |
|
51 |
+
if sources == ["ABC"]:
|
52 |
+
vectorstore = vectorstores["ABC"]
|
53 |
+
else:
|
54 |
+
vectorstore = vectorstores["XYZ"]
|
55 |
|
56 |
+
# get context
|
57 |
+
context_retrieved_lst = []
|
58 |
+
question_lst= [query]
|
59 |
+
for question in question_lst:
|
60 |
+
retriever = vectorstore.as_retriever(
|
61 |
+
search_type="similarity",
|
62 |
+
search_kwargs={"k": 1})
|
63 |
+
|
64 |
+
context_retrieved = retriever.get_relevant_documents(question)
|
65 |
+
|
66 |
+
def format_docs(docs):
|
67 |
+
return "\n\n".join(doc.page_content for doc in docs)
|
68 |
+
|
69 |
+
context_retrieved_formatted = format_docs(context_retrieved)
|
70 |
+
context_retrieved_lst.append(context_retrieved_formatted)
|
71 |
+
|
72 |
+
|
73 |
+
# get prompt
|
74 |
+
prompt = ChatPromptTemplate.from_template(answer_prompt_template)
|
75 |
+
|
76 |
+
# get llm
|
77 |
+
llm_qa = HuggingFaceEndpoint(
|
78 |
+
endpoint_url= "https://fesg9gjsfde5yfr4.us-east-1.aws.endpoints.huggingface.cloud",
|
79 |
+
task="text-generation",
|
80 |
+
huggingfacehub_api_token=HF_token,
|
81 |
+
model_kwargs={})
|
82 |
+
|
83 |
+
# create rag chain
|
84 |
+
chain = prompt | llm_qa | StrOutputParser()
|
85 |
+
# get answers
|
86 |
+
answer_lst = []
|
87 |
+
for question, context in zip(question_list , context_retrieved_lst):
|
88 |
+
answer = chain.invoke({"context": context, "question": question,'audience':audience_prompt, 'language':'english'})
|
89 |
+
answer_lst.append(answer)
|
90 |
+
docs_html = []
|
91 |
+
for i, d in enumerate(context_retrieved, 1):
|
92 |
+
docs_html.append(make_html_source(d, i))
|
93 |
+
docs_html = "".join(docs_html)
|
94 |
+
|
95 |
+
previous_answer = history[-1][1]
|
96 |
+
previous_answer = previous_answer if previous_answer is not None else ""
|
97 |
+
answer_yet = previous_answer + answer_lst[0]
|
98 |
+
answer_yet = parse_output_llm_with_sources(answer_yet)
|
99 |
+
history[-1] = (query,answer_yet)
|
100 |
+
|
101 |
+
history = [tuple(x) for x in history]
|
102 |
|
103 |
yield history,docs_html,output_query,output_language
|
104 |
|
105 |
+
def make_html_source(source,i):
|
106 |
+
meta = source.metadata
|
107 |
+
# content = source.page_content.split(":",1)[1].strip()
|
108 |
+
content = source.page_content.strip()
|
109 |
+
|
110 |
+
toc_levels = []
|
111 |
+
for j in range(2):
|
112 |
+
level = meta[f"toc_level{j}"]
|
113 |
+
if level != "N/A":
|
114 |
+
toc_levels.append(level)
|
115 |
+
else:
|
116 |
+
break
|
117 |
+
toc_levels = " > ".join(toc_levels)
|
118 |
+
|
119 |
+
if len(toc_levels) > 0:
|
120 |
+
name = f"<b>{toc_levels}</b><br/>{meta['name']}"
|
121 |
+
else:
|
122 |
+
name = meta['name']
|
123 |
+
|
124 |
+
if meta["chunk_type"] == "text":
|
125 |
+
|
126 |
+
card = f"""
|
127 |
+
<div class="card" id="doc{i}">
|
128 |
+
<div class="card-content">
|
129 |
+
<h2>Doc {i} - {meta['short_name']} - Page {int(meta['page_number'])}</h2>
|
130 |
+
<p>{content}</p>
|
131 |
+
</div>
|
132 |
+
<div class="card-footer">
|
133 |
+
<span>{name}</span>
|
134 |
+
<a href="{meta['url']}#page={int(meta['page_number'])}" target="_blank" class="pdf-link">
|
135 |
+
<span role="img" aria-label="Open PDF">🔗</span>
|
136 |
+
</a>
|
137 |
+
</div>
|
138 |
+
</div>
|
139 |
+
"""
|
140 |
+
return card
|
141 |
# --------------------------------------------------------------------
|
142 |
# Gradio
|
143 |
# --------------------------------------------------------------------
|