Update app.py
Browse files
app.py
CHANGED
@@ -1,6 +1,5 @@
|
|
1 |
import os
|
2 |
import shutil
|
3 |
-
import time
|
4 |
import gradio as gr
|
5 |
import qdrant_client
|
6 |
from getpass import getpass
|
@@ -34,66 +33,37 @@ client = None
|
|
34 |
vector_store = None
|
35 |
storage_context = None
|
36 |
|
37 |
-
# Define a persistent collection name.
|
38 |
-
collection_name = "paper"
|
39 |
-
|
40 |
-
# Use a persistent folder to store uploaded files.
|
41 |
-
upload_dir = "uploaded_files"
|
42 |
-
if not os.path.exists(upload_dir):
|
43 |
-
os.makedirs(upload_dir)
|
44 |
-
# We do not clear the folder to keep previously uploaded files.
|
45 |
-
|
46 |
# -------------------------------------------------------
|
47 |
-
# Function to process uploaded files and
|
48 |
# -------------------------------------------------------
|
49 |
def process_upload(files):
|
50 |
"""
|
51 |
-
Accepts a list of uploaded file paths, saves them to a
|
52 |
-
loads
|
53 |
"""
|
54 |
-
|
55 |
-
|
56 |
-
|
57 |
-
|
|
|
|
|
|
|
|
|
|
|
58 |
for file_path in files:
|
59 |
file_name = os.path.basename(file_path)
|
60 |
dest = os.path.join(upload_dir, file_name)
|
61 |
-
|
62 |
-
shutil.copy(file_path, dest)
|
63 |
-
new_file_paths.append(dest)
|
64 |
|
65 |
-
#
|
66 |
-
|
67 |
-
return "No new documents to add."
|
68 |
-
|
69 |
-
# Load only the new documents.
|
70 |
-
new_documents = SimpleDirectoryReader(input_files=new_file_paths).load_data()
|
71 |
-
|
72 |
-
# Initialize a persistent Qdrant client.
|
73 |
-
client = qdrant_client.QdrantClient(
|
74 |
-
path="./qdrant_db",
|
75 |
-
prefer_grpc=True
|
76 |
-
)
|
77 |
|
78 |
-
#
|
79 |
-
|
80 |
-
|
81 |
-
if collection_name not in existing_collections:
|
82 |
-
client.create_collection(
|
83 |
-
collection_name=collection_name,
|
84 |
-
vectors_config={
|
85 |
-
"text-dense": models.VectorParams(
|
86 |
-
size=1536, # text-embedding-ada-002 produces 1536-dimensional vectors.
|
87 |
-
distance=models.Distance.COSINE
|
88 |
-
)
|
89 |
-
}
|
90 |
-
)
|
91 |
-
# Wait briefly for the collection creation to complete.
|
92 |
-
time.sleep(1)
|
93 |
|
94 |
-
# Initialize (or re-use) the vector store.
|
95 |
vector_store = QdrantVectorStore(
|
96 |
-
collection_name=
|
97 |
client=client,
|
98 |
enable_hybrid=True,
|
99 |
batch_size=20,
|
@@ -101,19 +71,12 @@ def process_upload(files):
|
|
101 |
|
102 |
storage_context = StorageContext.from_defaults(vector_store=vector_store)
|
103 |
|
104 |
-
|
105 |
-
if index is None:
|
106 |
-
# Load all documents from the persistent folder.
|
107 |
-
index = VectorStoreIndex.from_documents(
|
108 |
-
SimpleDirectoryReader(upload_dir).load_data(),
|
109 |
-
storage_context=storage_context
|
110 |
-
)
|
111 |
-
else:
|
112 |
-
index.insert_documents(new_documents)
|
113 |
|
114 |
-
# Reinitialize query and chat engines to reflect updates.
|
115 |
query_engine = index.as_query_engine(vector_store_query_mode="hybrid")
|
|
|
116 |
memory = ChatMemoryBuffer.from_defaults(token_limit=3000)
|
|
|
117 |
chat_engine = index.as_chat_engine(
|
118 |
chat_mode="context",
|
119 |
memory=memory,
|
@@ -123,26 +86,32 @@ def process_upload(files):
|
|
123 |
),
|
124 |
)
|
125 |
|
126 |
-
return "Documents uploaded and index
|
127 |
|
128 |
# -------------------------------------------------------
|
129 |
# Chat function that uses the built chat engine.
|
130 |
# -------------------------------------------------------
|
131 |
def chat_with_ai(user_input, chat_history):
|
132 |
global chat_engine
|
|
|
133 |
if chat_engine is None:
|
134 |
return chat_history, "Please upload documents first."
|
135 |
|
136 |
response = chat_engine.chat(user_input)
|
137 |
references = response.source_nodes
|
138 |
-
ref = []
|
|
|
|
|
139 |
for node in references:
|
140 |
file_name = node.metadata.get('file_name')
|
141 |
if file_name and file_name not in ref:
|
142 |
ref.append(file_name)
|
143 |
|
144 |
complete_response = str(response) + "\n\n"
|
145 |
-
|
|
|
|
|
|
|
146 |
return chat_history, ""
|
147 |
|
148 |
# -------------------------------------------------------
|
@@ -161,6 +130,7 @@ def gradio_interface():
|
|
161 |
# Use Tabs to separate the file upload and chat interfaces.
|
162 |
with gr.Tab("Upload Documents"):
|
163 |
gr.Markdown("Upload PDF, Excel, CSV, DOC/DOCX, or TXT files below:")
|
|
|
164 |
file_upload = gr.File(
|
165 |
label="Upload Files",
|
166 |
file_count="multiple",
|
|
|
1 |
import os
|
2 |
import shutil
|
|
|
3 |
import gradio as gr
|
4 |
import qdrant_client
|
5 |
from getpass import getpass
|
|
|
33 |
vector_store = None
|
34 |
storage_context = None
|
35 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
36 |
# -------------------------------------------------------
|
37 |
+
# Function to process uploaded files and build the index.
|
38 |
# -------------------------------------------------------
|
39 |
def process_upload(files):
|
40 |
"""
|
41 |
+
Accepts a list of uploaded file paths, saves them to a local folder,
|
42 |
+
loads them as documents, and builds the vector index and chat engine.
|
43 |
"""
|
44 |
+
upload_dir = "uploaded_files"
|
45 |
+
if not os.path.exists(upload_dir):
|
46 |
+
os.makedirs(upload_dir)
|
47 |
+
else:
|
48 |
+
# Clear any existing files in the folder.
|
49 |
+
for f in os.listdir(upload_dir):
|
50 |
+
os.remove(os.path.join(upload_dir, f))
|
51 |
+
|
52 |
+
# 'files' is a list of file paths (Gradio's File component with type="file")
|
53 |
for file_path in files:
|
54 |
file_name = os.path.basename(file_path)
|
55 |
dest = os.path.join(upload_dir, file_name)
|
56 |
+
shutil.copy(file_path, dest)
|
|
|
|
|
57 |
|
58 |
+
# Load documents from the saved folder.
|
59 |
+
documents = SimpleDirectoryReader(upload_dir).load_data()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
60 |
|
61 |
+
# Build the index and chat engine using Qdrant as the vector store.
|
62 |
+
global client, vector_store, storage_context, index, query_engine, memory, chat_engine
|
63 |
+
client = qdrant_client.QdrantClient(location=":memory:")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
64 |
|
|
|
65 |
vector_store = QdrantVectorStore(
|
66 |
+
collection_name="paper",
|
67 |
client=client,
|
68 |
enable_hybrid=True,
|
69 |
batch_size=20,
|
|
|
71 |
|
72 |
storage_context = StorageContext.from_defaults(vector_store=vector_store)
|
73 |
|
74 |
+
index = VectorStoreIndex.from_documents(documents, storage_context=storage_context)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
75 |
|
|
|
76 |
query_engine = index.as_query_engine(vector_store_query_mode="hybrid")
|
77 |
+
|
78 |
memory = ChatMemoryBuffer.from_defaults(token_limit=3000)
|
79 |
+
|
80 |
chat_engine = index.as_chat_engine(
|
81 |
chat_mode="context",
|
82 |
memory=memory,
|
|
|
86 |
),
|
87 |
)
|
88 |
|
89 |
+
return "Documents uploaded and index built successfully!"
|
90 |
|
91 |
# -------------------------------------------------------
|
92 |
# Chat function that uses the built chat engine.
|
93 |
# -------------------------------------------------------
|
94 |
def chat_with_ai(user_input, chat_history):
|
95 |
global chat_engine
|
96 |
+
# Check if the chat engine is initialized.
|
97 |
if chat_engine is None:
|
98 |
return chat_history, "Please upload documents first."
|
99 |
|
100 |
response = chat_engine.chat(user_input)
|
101 |
references = response.source_nodes
|
102 |
+
ref, pages = [], []
|
103 |
+
|
104 |
+
# Extract file names from the source nodes (if available)
|
105 |
for node in references:
|
106 |
file_name = node.metadata.get('file_name')
|
107 |
if file_name and file_name not in ref:
|
108 |
ref.append(file_name)
|
109 |
|
110 |
complete_response = str(response) + "\n\n"
|
111 |
+
if ref or pages:
|
112 |
+
chat_history.append((user_input, complete_response))
|
113 |
+
else:
|
114 |
+
chat_history.append((user_input, str(response)))
|
115 |
return chat_history, ""
|
116 |
|
117 |
# -------------------------------------------------------
|
|
|
130 |
# Use Tabs to separate the file upload and chat interfaces.
|
131 |
with gr.Tab("Upload Documents"):
|
132 |
gr.Markdown("Upload PDF, Excel, CSV, DOC/DOCX, or TXT files below:")
|
133 |
+
# The file upload widget: we specify allowed file types.
|
134 |
file_upload = gr.File(
|
135 |
label="Upload Files",
|
136 |
file_count="multiple",
|