eikarna
commited on
Commit
·
33009e3
1
Parent(s):
d9760ae
Revert to non-RAG
Browse files- app.py.bak +264 -0
app.py.bak
ADDED
|
@@ -0,0 +1,264 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import streamlit as st
|
| 2 |
+
import requests
|
| 3 |
+
import logging
|
| 4 |
+
import time
|
| 5 |
+
from typing import Dict, Any, Optional, List
|
| 6 |
+
import os
|
| 7 |
+
from PIL import Image
|
| 8 |
+
import pytesseract
|
| 9 |
+
import fitz # PyMuPDF
|
| 10 |
+
from io import BytesIO
|
| 11 |
+
import hashlib
|
| 12 |
+
from sentence_transformers import SentenceTransformer
|
| 13 |
+
import numpy as np
|
| 14 |
+
from pathlib import Path
|
| 15 |
+
import pickle
|
| 16 |
+
import tempfile
|
| 17 |
+
|
| 18 |
+
# Configure logging
|
| 19 |
+
logging.basicConfig(
|
| 20 |
+
level=logging.INFO,
|
| 21 |
+
format='%(asctime)s - %(name)s - %(levelname)s - %(message)s'
|
| 22 |
+
)
|
| 23 |
+
logger = logging.getLogger(__name__)
|
| 24 |
+
|
| 25 |
+
# Initialize SBERT model for embeddings
|
| 26 |
+
@st.cache_resource
|
| 27 |
+
def load_embedding_model():
|
| 28 |
+
return SentenceTransformer('all-MiniLM-L6-v2')
|
| 29 |
+
|
| 30 |
+
# Modified Vector Store Class
|
| 31 |
+
class SimpleVectorStore:
|
| 32 |
+
def __init__(self):
|
| 33 |
+
self.documents = []
|
| 34 |
+
self.embeddings = []
|
| 35 |
+
|
| 36 |
+
def add_document(self, text: str, embedding: np.ndarray):
|
| 37 |
+
self.documents.append(text)
|
| 38 |
+
self.embeddings.append(embedding)
|
| 39 |
+
|
| 40 |
+
def search(self, query_embedding: np.ndarray, top_k: int = 3) -> List[str]:
|
| 41 |
+
if not self.embeddings:
|
| 42 |
+
return []
|
| 43 |
+
|
| 44 |
+
similarities = np.dot(self.embeddings, query_embedding)
|
| 45 |
+
top_indices = np.argsort(similarities)[-top_k:][::-1]
|
| 46 |
+
return [self.documents[i] for i in top_indices]
|
| 47 |
+
|
| 48 |
+
# Document processing functions
|
| 49 |
+
def process_text(text: str) -> List[str]:
|
| 50 |
+
"""Split text into chunks."""
|
| 51 |
+
# Simple splitting by sentences (can be improved with better chunking)
|
| 52 |
+
chunks = text.split('. ')
|
| 53 |
+
return [chunk + '.' for chunk in chunks if chunk]
|
| 54 |
+
|
| 55 |
+
def process_image(image) -> str:
|
| 56 |
+
"""Extract text from image using OCR."""
|
| 57 |
+
try:
|
| 58 |
+
text = pytesseract.image_to_string(image)
|
| 59 |
+
return text
|
| 60 |
+
except Exception as e:
|
| 61 |
+
logger.error(f"Error processing image: {str(e)}")
|
| 62 |
+
return ""
|
| 63 |
+
|
| 64 |
+
def process_pdf(pdf_file) -> str:
|
| 65 |
+
"""Extract text from PDF."""
|
| 66 |
+
try:
|
| 67 |
+
with tempfile.NamedTemporaryFile(delete=False) as tmp_file:
|
| 68 |
+
tmp_file.write(pdf_file.read())
|
| 69 |
+
tmp_file.flush()
|
| 70 |
+
|
| 71 |
+
doc = fitz.open(tmp_file.name)
|
| 72 |
+
text = ""
|
| 73 |
+
for page in doc:
|
| 74 |
+
text += page.get_text()
|
| 75 |
+
doc.close()
|
| 76 |
+
os.unlink(tmp_file.name)
|
| 77 |
+
return text
|
| 78 |
+
except Exception as e:
|
| 79 |
+
logger.error(f"Error processing PDF: {str(e)}")
|
| 80 |
+
return ""
|
| 81 |
+
|
| 82 |
+
# Initialize session state
|
| 83 |
+
if "messages" not in st.session_state:
|
| 84 |
+
st.session_state.messages = []
|
| 85 |
+
if "request_timestamps" not in st.session_state:
|
| 86 |
+
st.session_state.request_timestamps = []
|
| 87 |
+
if "vector_store" not in st.session_state:
|
| 88 |
+
st.session_state.vector_store = SimpleVectorStore()
|
| 89 |
+
|
| 90 |
+
# Rate limiting configuration
|
| 91 |
+
RATE_LIMIT_PERIOD = 60
|
| 92 |
+
MAX_REQUESTS_PER_PERIOD = 30
|
| 93 |
+
|
| 94 |
+
def check_rate_limit() -> bool:
|
| 95 |
+
"""Check if we're within rate limits."""
|
| 96 |
+
current_time = time.time()
|
| 97 |
+
st.session_state.request_timestamps = [
|
| 98 |
+
ts for ts in st.session_state.request_timestamps
|
| 99 |
+
if current_time - ts < RATE_LIMIT_PERIOD
|
| 100 |
+
]
|
| 101 |
+
|
| 102 |
+
if len(st.session_state.request_timestamps) >= MAX_REQUESTS_PER_PERIOD:
|
| 103 |
+
return False
|
| 104 |
+
|
| 105 |
+
st.session_state.request_timestamps.append(current_time)
|
| 106 |
+
return True
|
| 107 |
+
|
| 108 |
+
def query(payload: Dict[str, Any], api_url: str) -> Optional[Dict[str, Any]]:
|
| 109 |
+
"""Query the Hugging Face API with error handling and rate limiting."""
|
| 110 |
+
if not check_rate_limit():
|
| 111 |
+
raise Exception(f"Rate limit exceeded. Please wait {RATE_LIMIT_PERIOD} seconds.")
|
| 112 |
+
|
| 113 |
+
try:
|
| 114 |
+
headers = {"Authorization": f"Bearer {st.secrets['HF_TOKEN']}"}
|
| 115 |
+
response = requests.post(api_url, headers=headers, json=payload, timeout=30)
|
| 116 |
+
|
| 117 |
+
if response.status_code == 429:
|
| 118 |
+
raise Exception("Too many requests. Please try again later.")
|
| 119 |
+
|
| 120 |
+
response.raise_for_status()
|
| 121 |
+
print(response.request.url)
|
| 122 |
+
print(response.request.headers)
|
| 123 |
+
print(response.request.body)
|
| 124 |
+
print(response)
|
| 125 |
+
return response.json()
|
| 126 |
+
except requests.exceptions.JSONDecodeError as e:
|
| 127 |
+
logger.error(f"API request failed: {str(e)}")
|
| 128 |
+
raise
|
| 129 |
+
|
| 130 |
+
# Enhanced response validation
|
| 131 |
+
def process_response(response: Dict[str, Any]) -> str:
|
| 132 |
+
if not isinstance(response, list) or not response:
|
| 133 |
+
raise ValueError("Invalid response format")
|
| 134 |
+
|
| 135 |
+
if 'generated_text' not in response[0]:
|
| 136 |
+
raise ValueError("Unexpected response structure")
|
| 137 |
+
|
| 138 |
+
text = response[0]['generated_text'].strip()
|
| 139 |
+
|
| 140 |
+
# Page configuration
|
| 141 |
+
st.set_page_config(
|
| 142 |
+
page_title="RAG-Enabled DeepSeek Chatbot",
|
| 143 |
+
page_icon="🤖",
|
| 144 |
+
layout="wide"
|
| 145 |
+
)
|
| 146 |
+
|
| 147 |
+
# Sidebar configuration
|
| 148 |
+
with st.sidebar:
|
| 149 |
+
st.header("Model Configuration")
|
| 150 |
+
st.markdown("[Get HuggingFace Token](https://huggingface.co/settings/tokens)")
|
| 151 |
+
|
| 152 |
+
model_options = [
|
| 153 |
+
"deepseek-ai/DeepSeek-R1-Distill-Qwen-32B",
|
| 154 |
+
]
|
| 155 |
+
selected_model = st.selectbox("Select Model", model_options, index=0)
|
| 156 |
+
|
| 157 |
+
system_message = st.text_area(
|
| 158 |
+
"System Message",
|
| 159 |
+
value="You are a friendly chatbot with RAG capabilities. Use the provided context to answer questions accurately. If the context doesn't contain relevant information, say so.",
|
| 160 |
+
height=100
|
| 161 |
+
)
|
| 162 |
+
|
| 163 |
+
max_tokens = st.slider("Max Tokens", 10, 4000, 100)
|
| 164 |
+
temperature = st.slider("Temperature", 0.1, 4.0, 0.3)
|
| 165 |
+
top_p = st.slider("Top-p", 0.1, 1.0, 0.6)
|
| 166 |
+
|
| 167 |
+
# File upload section
|
| 168 |
+
st.header("Upload Knowledge Base")
|
| 169 |
+
uploaded_files = st.file_uploader(
|
| 170 |
+
"Upload files (PDF, Images, Text)",
|
| 171 |
+
type=['pdf', 'png', 'jpg', 'jpeg', 'txt'],
|
| 172 |
+
accept_multiple_files=True
|
| 173 |
+
)
|
| 174 |
+
|
| 175 |
+
# Process uploaded files
|
| 176 |
+
if uploaded_files:
|
| 177 |
+
embedding_model = load_embedding_model()
|
| 178 |
+
|
| 179 |
+
for file in uploaded_files:
|
| 180 |
+
try:
|
| 181 |
+
if file.type == "application/pdf":
|
| 182 |
+
text = process_pdf(file)
|
| 183 |
+
elif file.type.startswith("image/"):
|
| 184 |
+
image = Image.open(file)
|
| 185 |
+
text = process_image(image)
|
| 186 |
+
else: # text files
|
| 187 |
+
text = file.getvalue().decode()
|
| 188 |
+
|
| 189 |
+
chunks = process_text(text)
|
| 190 |
+
for chunk in chunks:
|
| 191 |
+
embedding = embedding_model.encode(chunk)
|
| 192 |
+
st.session_state.vector_store.add_document(chunk, embedding)
|
| 193 |
+
|
| 194 |
+
st.sidebar.success(f"Successfully processed {file.name}")
|
| 195 |
+
except Exception as e:
|
| 196 |
+
st.sidebar.error(f"Error processing {file.name}: {str(e)}")
|
| 197 |
+
|
| 198 |
+
# Main chat interface
|
| 199 |
+
st.title("🤖 RAG-Enabled DeepSeek Chatbot")
|
| 200 |
+
st.caption("Upload documents in the sidebar to enhance the chatbot's knowledge")
|
| 201 |
+
|
| 202 |
+
# Display chat history
|
| 203 |
+
for message in st.session_state.messages:
|
| 204 |
+
with st.chat_message(message["role"]):
|
| 205 |
+
st.markdown(message["content"])
|
| 206 |
+
|
| 207 |
+
# Handle user input
|
| 208 |
+
if prompt := st.chat_input("Type your message..."):
|
| 209 |
+
# Display user message
|
| 210 |
+
st.session_state.messages.append({"role": "user", "content": prompt})
|
| 211 |
+
with st.chat_message("user"):
|
| 212 |
+
st.markdown(prompt)
|
| 213 |
+
|
| 214 |
+
try:
|
| 215 |
+
with st.spinner("Generating response..."):
|
| 216 |
+
embedding_model = load_embedding_model()
|
| 217 |
+
query_embedding = embedding_model.encode(prompt)
|
| 218 |
+
relevant_contexts = st.session_state.vector_store.search(query_embedding)
|
| 219 |
+
|
| 220 |
+
# Dynamic context handling
|
| 221 |
+
context_text = "\n".join(relevant_contexts) if relevant_contexts else ""
|
| 222 |
+
system_msg = (
|
| 223 |
+
f"{system_message} Use the provided context to answer accurately."
|
| 224 |
+
if context_text
|
| 225 |
+
else system_message
|
| 226 |
+
)
|
| 227 |
+
|
| 228 |
+
# Format for DeepSeek model
|
| 229 |
+
full_prompt = f"""<|beginofutterance|>System: {system_msg}
|
| 230 |
+
{context_text if context_text else ''}
|
| 231 |
+
<|endofutterance|>
|
| 232 |
+
<|beginofutterance|>User: {prompt}<|endofutterance|>
|
| 233 |
+
<|beginofutterance|>Assistant:"""
|
| 234 |
+
|
| 235 |
+
payload = {
|
| 236 |
+
"inputs": full_prompt,
|
| 237 |
+
"parameters": {
|
| 238 |
+
"max_new_tokens": max_tokens,
|
| 239 |
+
"temperature": temperature,
|
| 240 |
+
"top_p": top_p,
|
| 241 |
+
"return_full_text": False
|
| 242 |
+
}
|
| 243 |
+
}
|
| 244 |
+
|
| 245 |
+
api_url = f"https://api-inference.huggingface.co/models/{selected_model}"
|
| 246 |
+
|
| 247 |
+
# Get and process response
|
| 248 |
+
output = query(payload, api_url)
|
| 249 |
+
if output:
|
| 250 |
+
response_text = process_response(output)
|
| 251 |
+
|
| 252 |
+
# Display assistant response
|
| 253 |
+
with st.chat_message("assistant"):
|
| 254 |
+
st.markdown(response_text)
|
| 255 |
+
|
| 256 |
+
# Update chat history
|
| 257 |
+
st.session_state.messages.append({
|
| 258 |
+
"role": "assistant",
|
| 259 |
+
"content": response_text
|
| 260 |
+
})
|
| 261 |
+
|
| 262 |
+
except Exception as e:
|
| 263 |
+
logger.error(f"Error: {str(e)}", exc_info=True)
|
| 264 |
+
st.error(f"Error: {str(e)}")
|