Spaces:
Sleeping
Sleeping
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,189 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import os
|
3 |
+
import numpy as np
|
4 |
+
import faiss
|
5 |
+
from mistralai import Mistral
|
6 |
+
|
7 |
+
api_key = os.getenv("MISTRAL_API_KEY")
|
8 |
+
client = Mistral(api_key=api_key)
|
9 |
+
|
10 |
+
# =============================================================================
|
11 |
+
# BASIC CHAT UI (Gradio Version)
|
12 |
+
# =============================================================================
|
13 |
+
|
14 |
+
def run_mistral_basic(message, history):
|
15 |
+
"""Basic chat function for Gradio ChatInterface"""
|
16 |
+
messages = [{"role": "user", "content": message}]
|
17 |
+
chat_response = client.chat.complete(
|
18 |
+
model="mistral-large-latest",
|
19 |
+
messages=messages
|
20 |
+
)
|
21 |
+
return chat_response.choices[0].message.content
|
22 |
+
|
23 |
+
# Create basic chat interface
|
24 |
+
basic_chat = gr.ChatInterface(
|
25 |
+
fn=run_mistral_basic,
|
26 |
+
title="Basic Mistral Chat",
|
27 |
+
description="Chat with Mistral AI"
|
28 |
+
)
|
29 |
+
|
30 |
+
# =============================================================================
|
31 |
+
# RAG UI (Gradio Version)
|
32 |
+
# =============================================================================
|
33 |
+
|
34 |
+
# Global variable to store processed document
|
35 |
+
processed_chunks = None
|
36 |
+
faiss_index = None
|
37 |
+
|
38 |
+
def get_text_embedding(input_text):
|
39 |
+
"""Get embeddings from Mistral"""
|
40 |
+
embeddings_batch_response = client.embeddings.create(
|
41 |
+
model="mistral-embed",
|
42 |
+
inputs=[input_text]
|
43 |
+
)
|
44 |
+
return embeddings_batch_response.data[0].embedding
|
45 |
+
|
46 |
+
def process_document(file):
|
47 |
+
"""Process uploaded document and create FAISS index"""
|
48 |
+
global processed_chunks, faiss_index
|
49 |
+
|
50 |
+
if file is None:
|
51 |
+
return "Please upload a text file first."
|
52 |
+
|
53 |
+
try:
|
54 |
+
# Read the file
|
55 |
+
with open(file.name, 'r', encoding='utf-8') as f:
|
56 |
+
text = f.read()
|
57 |
+
|
58 |
+
# Split document into chunks
|
59 |
+
chunk_size = 2048
|
60 |
+
processed_chunks = [text[i:i + chunk_size] for i in range(0, len(text), chunk_size)]
|
61 |
+
|
62 |
+
# Create embeddings and FAISS index
|
63 |
+
text_embeddings = np.array([get_text_embedding(chunk) for chunk in processed_chunks])
|
64 |
+
d = text_embeddings.shape[1]
|
65 |
+
faiss_index = faiss.IndexFlatL2(d)
|
66 |
+
faiss_index.add(text_embeddings.astype(np.float32))
|
67 |
+
|
68 |
+
return f"Document processed successfully! Split into {len(processed_chunks)} chunks."
|
69 |
+
|
70 |
+
except Exception as e:
|
71 |
+
return f"Error processing document: {str(e)}"
|
72 |
+
|
73 |
+
def rag_chat(message, history):
|
74 |
+
"""RAG chat function for Gradio"""
|
75 |
+
global processed_chunks, faiss_index
|
76 |
+
|
77 |
+
if processed_chunks is None or faiss_index is None:
|
78 |
+
return "Please upload and process a document first."
|
79 |
+
|
80 |
+
try:
|
81 |
+
# Create prompt template
|
82 |
+
prompt_template = """
|
83 |
+
Context information is below.
|
84 |
+
---------------------
|
85 |
+
{retrieved_chunk}
|
86 |
+
---------------------
|
87 |
+
Given the context information and not prior knowledge, answer the query.
|
88 |
+
Query: {question}
|
89 |
+
Answer:
|
90 |
+
"""
|
91 |
+
|
92 |
+
# Get question embedding
|
93 |
+
question_embedding = np.array([get_text_embedding(message)])
|
94 |
+
|
95 |
+
# Search for similar chunks
|
96 |
+
D, I = faiss_index.search(question_embedding.astype(np.float32), k=2)
|
97 |
+
retrieved_chunks = [processed_chunks[i] for i in I.tolist()[0]]
|
98 |
+
|
99 |
+
# Generate response
|
100 |
+
prompt = prompt_template.format(
|
101 |
+
retrieved_chunk=retrieved_chunks,
|
102 |
+
question=message
|
103 |
+
)
|
104 |
+
|
105 |
+
messages = [{"role": "user", "content": prompt}]
|
106 |
+
chat_response = client.chat.complete(
|
107 |
+
model="mistral-large-latest",
|
108 |
+
messages=messages
|
109 |
+
)
|
110 |
+
|
111 |
+
return chat_response.choices[0].message.content
|
112 |
+
|
113 |
+
except Exception as e:
|
114 |
+
return f"Error generating response: {str(e)}"
|
115 |
+
|
116 |
+
# =============================================================================
|
117 |
+
# GRADIO INTERFACES
|
118 |
+
# =============================================================================
|
119 |
+
|
120 |
+
# Create RAG interface with file upload
|
121 |
+
with gr.Blocks(title="RAG Chat with Mistral") as rag_interface:
|
122 |
+
gr.Markdown("# RAG Chat Interface")
|
123 |
+
gr.Markdown("Upload a text file and chat with its content!")
|
124 |
+
|
125 |
+
with gr.Row():
|
126 |
+
file_upload = gr.File(
|
127 |
+
label="Upload Text File",
|
128 |
+
file_types=[".txt"],
|
129 |
+
type="filepath"
|
130 |
+
)
|
131 |
+
process_btn = gr.Button("Process Document", variant="primary")
|
132 |
+
|
133 |
+
process_status = gr.Textbox(
|
134 |
+
label="Processing Status",
|
135 |
+
interactive=False,
|
136 |
+
placeholder="Upload a file and click 'Process Document'"
|
137 |
+
)
|
138 |
+
|
139 |
+
# Chat interface
|
140 |
+
chatbot = gr.Chatbot(label="RAG Chat")
|
141 |
+
msg = gr.Textbox(
|
142 |
+
label="Your Message",
|
143 |
+
placeholder="Ask questions about the uploaded document...",
|
144 |
+
lines=2
|
145 |
+
)
|
146 |
+
|
147 |
+
with gr.Row():
|
148 |
+
submit_btn = gr.Button("Send", variant="primary")
|
149 |
+
clear_btn = gr.Button("Clear Chat")
|
150 |
+
|
151 |
+
# Event handlers
|
152 |
+
process_btn.click(
|
153 |
+
process_document,
|
154 |
+
inputs=[file_upload],
|
155 |
+
outputs=[process_status]
|
156 |
+
)
|
157 |
+
|
158 |
+
def respond(message, chat_history):
|
159 |
+
if not message.strip():
|
160 |
+
return "", chat_history
|
161 |
+
|
162 |
+
# Add user message to history
|
163 |
+
chat_history.append([message, None])
|
164 |
+
|
165 |
+
# Get bot response
|
166 |
+
bot_response = rag_chat(message, chat_history)
|
167 |
+
|
168 |
+
# Add bot response to history
|
169 |
+
chat_history[-1][1] = bot_response
|
170 |
+
|
171 |
+
return "", chat_history
|
172 |
+
|
173 |
+
submit_btn.click(
|
174 |
+
respond,
|
175 |
+
inputs=[msg, chatbot],
|
176 |
+
outputs=[msg, chatbot]
|
177 |
+
)
|
178 |
+
|
179 |
+
msg.submit(
|
180 |
+
respond,
|
181 |
+
inputs=[msg, chatbot],
|
182 |
+
outputs=[msg, chatbot]
|
183 |
+
)
|
184 |
+
|
185 |
+
clear_btn.click(lambda: ([], ""), outputs=[chatbot, msg])
|
186 |
+
|
187 |
+
if __name__ == "__main__":
|
188 |
+
rag_interface.launch(share=True)
|
189 |
+
|