Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,84 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import PyPDF2
|
3 |
+
from transformers import AutoTokenizer, AutoModel
|
4 |
+
import torch
|
5 |
+
import weaviate
|
6 |
+
import cohere
|
7 |
+
|
8 |
+
auth_config = weaviate.AuthApiKey(api_key="16LRz5YwOtnq8ov51Lhg1UuAollpsMgspulV")
|
9 |
+
client = weaviate.Client(
|
10 |
+
url="https://wkoll9rds3orbu9fhzfr2a.c0.asia-southeast1.gcp.weaviate.cloud",
|
11 |
+
auth_client_secret=auth_config
|
12 |
+
)
|
13 |
+
cohere_client = cohere.Client("LEvCVeZkqZMW1aLYjxDqlstCzWi4Cvlt9PiysqT8")
|
14 |
+
|
15 |
+
def load_pdf(file):
|
16 |
+
reader = PyPDF2.PdfReader(file)
|
17 |
+
text = ''
|
18 |
+
for page in range(len(reader.pages)):
|
19 |
+
text += reader.pages[page].extract_text()
|
20 |
+
return text
|
21 |
+
|
22 |
+
tokenizer = AutoTokenizer.from_pretrained('sentence-transformers/all-MiniLM-L6-v2')
|
23 |
+
model = AutoModel.from_pretrained('sentence-transformers/all-MiniLM-L6-v2')
|
24 |
+
|
25 |
+
def get_embeddings(text):
|
26 |
+
inputs = tokenizer(text, return_tensors="pt", padding=True, truncation=True)
|
27 |
+
with torch.no_grad():
|
28 |
+
embeddings = model(**inputs).last_hidden_state.mean(dim=1).squeeze().cpu().numpy()
|
29 |
+
return embeddings
|
30 |
+
|
31 |
+
def upload_document_chunks(chunks):
|
32 |
+
for idx, chunk in enumerate(chunks):
|
33 |
+
embedding = get_embeddings(chunk)
|
34 |
+
client.data_object.create(
|
35 |
+
{"content": chunk},
|
36 |
+
"Document",
|
37 |
+
vector=embedding.tolist()
|
38 |
+
)
|
39 |
+
|
40 |
+
def query_answer(query):
|
41 |
+
query_embedding = get_embeddings(query)
|
42 |
+
result = client.query.get("Document", ["content"])\
|
43 |
+
.with_near_vector({"vector": query_embedding.tolist()})\
|
44 |
+
.with_limit(3)\
|
45 |
+
.do()
|
46 |
+
return result
|
47 |
+
|
48 |
+
def generate_response(context, query):
|
49 |
+
response = cohere_client.generate(
|
50 |
+
model='command',
|
51 |
+
prompt=f"Context: {context}\n\nQuestion: {query}?\nAnswer:",
|
52 |
+
max_tokens=100
|
53 |
+
)
|
54 |
+
return response.generations[0].text.strip()
|
55 |
+
|
56 |
+
def qa_pipeline(pdf_file, query):
|
57 |
+
document_text = load_pdf(pdf_file)
|
58 |
+
document_chunks = [document_text[i:i+500] for i in range(0, len(document_text), 500)]
|
59 |
+
|
60 |
+
upload_document_chunks(document_chunks)
|
61 |
+
|
62 |
+
response = query_answer(query)
|
63 |
+
context = ' '.join([doc['content'] for doc in response['data']['Get']['Document']])
|
64 |
+
|
65 |
+
answer = generate_response(context, query)
|
66 |
+
|
67 |
+
return context, answer
|
68 |
+
|
69 |
+
with gr.Blocks() as demo:
|
70 |
+
gr.Markdown("# Interactive QA Bot")
|
71 |
+
|
72 |
+
pdf_input = gr.File(label="Upload a PDF file", file_types=[".pdf"])
|
73 |
+
query_input = gr.Textbox(label="Ask a question")
|
74 |
+
|
75 |
+
doc_segments_output = gr.Textbox(label="Retrieved Document Segments")
|
76 |
+
answer_output = gr.Textbox(label="Answer")
|
77 |
+
|
78 |
+
gr.Button("Submit").click(
|
79 |
+
qa_pipeline,
|
80 |
+
inputs=[pdf_input, query_input],
|
81 |
+
outputs=[doc_segments_output, answer_output]
|
82 |
+
)
|
83 |
+
|
84 |
+
demo.launch()
|