Update src/streamlit_app.py
Browse files- src/streamlit_app.py +51 -72
src/streamlit_app.py
CHANGED
|
@@ -5,7 +5,7 @@ import streamlit as st
|
|
| 5 |
import torch
|
| 6 |
|
| 7 |
# ==========================================================
|
| 8 |
-
# β
Environment
|
| 9 |
# ==========================================================
|
| 10 |
print("CUDA available:", torch.cuda.is_available())
|
| 11 |
if torch.cuda.is_available():
|
|
@@ -13,20 +13,13 @@ if torch.cuda.is_available():
|
|
| 13 |
else:
|
| 14 |
print("Running on CPU")
|
| 15 |
|
| 16 |
-
# ==========================================================
|
| 17 |
-
# β
Page Configuration
|
| 18 |
-
# ==========================================================
|
| 19 |
st.set_page_config(page_title="Enterprise Knowledge Assistant", layout="wide")
|
| 20 |
|
| 21 |
# ==========================================================
|
| 22 |
-
# π§Ή Cache
|
| 23 |
# ==========================================================
|
| 24 |
def clean_cache(max_size_gb: float = 2.0):
|
| 25 |
-
folders = [
|
| 26 |
-
"/root/.cache/huggingface",
|
| 27 |
-
"/root/.cache/transformers",
|
| 28 |
-
"/root/.cache/torch",
|
| 29 |
-
]
|
| 30 |
total_deleted = 0.0
|
| 31 |
for folder in folders:
|
| 32 |
if os.path.exists(folder):
|
|
@@ -44,7 +37,7 @@ def clean_cache(max_size_gb: float = 2.0):
|
|
| 44 |
clean_cache()
|
| 45 |
|
| 46 |
# ==========================================================
|
| 47 |
-
# βοΈ Hugging Face Cache
|
| 48 |
# ==========================================================
|
| 49 |
CACHE_DIR = "/tmp/hf_cache"
|
| 50 |
os.makedirs(CACHE_DIR, exist_ok=True)
|
|
@@ -66,7 +59,6 @@ from qa import retrieve_chunks, generate_answer, cache_embeddings, embed_chunks,
|
|
| 66 |
# π§ Smart Suggestion Generator
|
| 67 |
# ==========================================================
|
| 68 |
def generate_dynamic_suggestions_from_toc(toc, chunks, doc_name="Document"):
|
| 69 |
-
"""Generate natural-sounding, concise professional questions from TOC + context."""
|
| 70 |
if not toc or not chunks:
|
| 71 |
return []
|
| 72 |
|
|
@@ -79,15 +71,17 @@ def generate_dynamic_suggestions_from_toc(toc, chunks, doc_name="Document"):
|
|
| 79 |
|
| 80 |
context_sample = " ".join(chunks[:3])[:4000]
|
| 81 |
prompt = f"""
|
| 82 |
-
You are
|
| 83 |
-
|
| 84 |
-
Use the table of contents and content sample below.
|
| 85 |
|
| 86 |
TABLE OF CONTENTS:
|
| 87 |
-
{chr(10).join(['- ' + t for t in titles[:
|
| 88 |
|
| 89 |
CONTENT SAMPLE:
|
| 90 |
{context_sample}
|
|
|
|
|
|
|
|
|
|
| 91 |
"""
|
| 92 |
|
| 93 |
try:
|
|
@@ -101,18 +95,7 @@ def generate_dynamic_suggestions_from_toc(toc, chunks, doc_name="Document"):
|
|
| 101 |
final.append(q)
|
| 102 |
return final[:7]
|
| 103 |
except Exception:
|
| 104 |
-
return [
|
| 105 |
-
"What is this document about?",
|
| 106 |
-
"How do I start using this process?",
|
| 107 |
-
"What configurations are needed?",
|
| 108 |
-
]
|
| 109 |
-
|
| 110 |
-
# ==========================================================
|
| 111 |
-
# π Paths
|
| 112 |
-
# ==========================================================
|
| 113 |
-
BASE_DIR = os.path.dirname(__file__)
|
| 114 |
-
LOGO_PATH = os.path.join(BASE_DIR, "logo.png")
|
| 115 |
-
SAMPLE_PATH = os.path.join(BASE_DIR, "sample.pdf")
|
| 116 |
|
| 117 |
# ==========================================================
|
| 118 |
# π₯οΈ Header
|
|
@@ -124,9 +107,6 @@ st.caption("Ask questions about SAP documentation and enterprise PDFs β powere
|
|
| 124 |
# π§ Sidebar
|
| 125 |
# ==========================================================
|
| 126 |
with st.sidebar:
|
| 127 |
-
if os.path.exists(LOGO_PATH):
|
| 128 |
-
st.image(LOGO_PATH, width=150)
|
| 129 |
-
|
| 130 |
if "reasoning_mode" not in st.session_state:
|
| 131 |
st.session_state.reasoning_mode = False
|
| 132 |
st.session_state.reasoning_mode = st.toggle(
|
|
@@ -148,28 +128,46 @@ with st.sidebar:
|
|
| 148 |
st.caption("β¨ Built by Shubham Sharma")
|
| 149 |
|
| 150 |
# ==========================================================
|
| 151 |
-
# π§Ύ Document Handling
|
| 152 |
# ==========================================================
|
| 153 |
text, chunks, index, embeddings, toc = None, None, None, None, None
|
| 154 |
|
| 155 |
-
#
|
| 156 |
-
|
| 157 |
-
|
| 158 |
-
|
| 159 |
-
|
| 160 |
-
|
| 161 |
-
|
| 162 |
-
|
| 163 |
-
|
| 164 |
-
|
| 165 |
-
|
| 166 |
-
|
| 167 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 168 |
if doc_choice == "-- Select --":
|
| 169 |
st.info("β¬
οΈ Please choose a document from the sidebar to begin.")
|
| 170 |
else:
|
| 171 |
if doc_choice == "Sample PDF":
|
| 172 |
-
temp_path =
|
| 173 |
st.success("π Using built-in Sample PDF")
|
| 174 |
else:
|
| 175 |
uploaded_file = st.file_uploader("π Upload your PDF", type="pdf")
|
|
@@ -193,10 +191,7 @@ else:
|
|
| 193 |
st.text_area("TOC Preview", toc_text, height=180)
|
| 194 |
query_suggestions = generate_dynamic_suggestions_from_toc(toc, chunks, os.path.basename(temp_path))
|
| 195 |
else:
|
| 196 |
-
query_suggestions = [
|
| 197 |
-
"What is this document about?",
|
| 198 |
-
"How do I start using this process?",
|
| 199 |
-
]
|
| 200 |
|
| 201 |
with st.spinner("βοΈ Preparing search index..."):
|
| 202 |
embeddings = cache_embeddings(os.path.basename(temp_path), chunks, embed_chunks)
|
|
@@ -211,34 +206,21 @@ else:
|
|
| 211 |
if query_suggestions:
|
| 212 |
st.markdown("#### π‘ Suggested Questions")
|
| 213 |
|
| 214 |
-
|
| 215 |
-
cols = st.columns(min(3, len(
|
| 216 |
|
| 217 |
-
for i, q in enumerate(
|
| 218 |
-
|
| 219 |
-
if col.button(f"π {q}", key=f"suggest_{i}"):
|
| 220 |
-
st.session_state.pending_query = q
|
| 221 |
-
st.experimental_rerun()
|
| 222 |
|
| 223 |
toggle_text = "Show less β²" if st.session_state.show_more else "Show more βΌ"
|
| 224 |
-
if st.button(toggle_text
|
| 225 |
st.session_state.show_more = not st.session_state.show_more
|
| 226 |
st.experimental_rerun()
|
| 227 |
|
| 228 |
-
|
| 229 |
-
if st.session_state.pending_query:
|
| 230 |
-
st.session_state.user_query_input = st.session_state.pending_query
|
| 231 |
-
st.session_state.pending_query = None
|
| 232 |
-
|
| 233 |
-
user_query = st.text_input(
|
| 234 |
-
"Type your question or pick one above:",
|
| 235 |
-
value=st.session_state.user_query_input,
|
| 236 |
-
key="user_query_input",
|
| 237 |
-
)
|
| 238 |
|
| 239 |
if user_query.strip():
|
| 240 |
st.caption("Mode: π§ Reasoning" if st.session_state.reasoning_mode else "Mode: π Strict Document")
|
| 241 |
-
|
| 242 |
with st.spinner("π Analyzing your document..."):
|
| 243 |
retrieved = retrieve_chunks(user_query, index, chunks, top_k=top_k, embeddings=embeddings)
|
| 244 |
answer = generate_answer(user_query, retrieved, reasoning_mode=st.session_state.reasoning_mode)
|
|
@@ -250,9 +232,6 @@ else:
|
|
| 250 |
for i, r in enumerate(retrieved, start=1):
|
| 251 |
st.markdown(f"**Chunk {i}:** {r}")
|
| 252 |
|
| 253 |
-
# ----------------------------------------------------------
|
| 254 |
-
# π Document Preview
|
| 255 |
-
# ----------------------------------------------------------
|
| 256 |
if chunks:
|
| 257 |
st.markdown("---")
|
| 258 |
st.subheader("π Document Preview")
|
|
|
|
| 5 |
import torch
|
| 6 |
|
| 7 |
# ==========================================================
|
| 8 |
+
# β
Environment Setup
|
| 9 |
# ==========================================================
|
| 10 |
print("CUDA available:", torch.cuda.is_available())
|
| 11 |
if torch.cuda.is_available():
|
|
|
|
| 13 |
else:
|
| 14 |
print("Running on CPU")
|
| 15 |
|
|
|
|
|
|
|
|
|
|
| 16 |
st.set_page_config(page_title="Enterprise Knowledge Assistant", layout="wide")
|
| 17 |
|
| 18 |
# ==========================================================
|
| 19 |
+
# π§Ή Cache Cleanup
|
| 20 |
# ==========================================================
|
| 21 |
def clean_cache(max_size_gb: float = 2.0):
|
| 22 |
+
folders = ["/root/.cache/huggingface", "/root/.cache/transformers", "/root/.cache/torch"]
|
|
|
|
|
|
|
|
|
|
|
|
|
| 23 |
total_deleted = 0.0
|
| 24 |
for folder in folders:
|
| 25 |
if os.path.exists(folder):
|
|
|
|
| 37 |
clean_cache()
|
| 38 |
|
| 39 |
# ==========================================================
|
| 40 |
+
# βοΈ Hugging Face Cache
|
| 41 |
# ==========================================================
|
| 42 |
CACHE_DIR = "/tmp/hf_cache"
|
| 43 |
os.makedirs(CACHE_DIR, exist_ok=True)
|
|
|
|
| 59 |
# π§ Smart Suggestion Generator
|
| 60 |
# ==========================================================
|
| 61 |
def generate_dynamic_suggestions_from_toc(toc, chunks, doc_name="Document"):
|
|
|
|
| 62 |
if not toc or not chunks:
|
| 63 |
return []
|
| 64 |
|
|
|
|
| 71 |
|
| 72 |
context_sample = " ".join(chunks[:3])[:4000]
|
| 73 |
prompt = f"""
|
| 74 |
+
You are generating intelligent, short, and context-aware questions from a document titled "{doc_name}".
|
| 75 |
+
Use the following Table of Contents and sample content for reference.
|
|
|
|
| 76 |
|
| 77 |
TABLE OF CONTENTS:
|
| 78 |
+
{chr(10).join(['- ' + t for t in titles[:8]])}
|
| 79 |
|
| 80 |
CONTENT SAMPLE:
|
| 81 |
{context_sample}
|
| 82 |
+
|
| 83 |
+
Generate 5β7 concise, professional, and relevant questions a user might ask about this document.
|
| 84 |
+
Each question should be under 20 words and directly based on the context.
|
| 85 |
"""
|
| 86 |
|
| 87 |
try:
|
|
|
|
| 95 |
final.append(q)
|
| 96 |
return final[:7]
|
| 97 |
except Exception:
|
| 98 |
+
return ["What is this document about?", "How do I start using this process?"]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 99 |
|
| 100 |
# ==========================================================
|
| 101 |
# π₯οΈ Header
|
|
|
|
| 107 |
# π§ Sidebar
|
| 108 |
# ==========================================================
|
| 109 |
with st.sidebar:
|
|
|
|
|
|
|
|
|
|
| 110 |
if "reasoning_mode" not in st.session_state:
|
| 111 |
st.session_state.reasoning_mode = False
|
| 112 |
st.session_state.reasoning_mode = st.toggle(
|
|
|
|
| 128 |
st.caption("β¨ Built by Shubham Sharma")
|
| 129 |
|
| 130 |
# ==========================================================
|
| 131 |
+
# π§Ύ Document Handling + Ask Section
|
| 132 |
# ==========================================================
|
| 133 |
text, chunks, index, embeddings, toc = None, None, None, None, None
|
| 134 |
|
| 135 |
+
# Style for chips
|
| 136 |
+
st.markdown("""
|
| 137 |
+
<style>
|
| 138 |
+
.suggest-chip {
|
| 139 |
+
background-color: #1f2937;
|
| 140 |
+
border: 1px solid #374151;
|
| 141 |
+
border-radius: 20px;
|
| 142 |
+
color: #f9fafb;
|
| 143 |
+
padding: 8px 14px;
|
| 144 |
+
cursor: pointer;
|
| 145 |
+
font-size: 14px;
|
| 146 |
+
transition: all 0.2s ease-in-out;
|
| 147 |
+
}
|
| 148 |
+
.suggest-chip:hover {
|
| 149 |
+
background-color: #2563eb;
|
| 150 |
+
border-color: #3b82f6;
|
| 151 |
+
color: #ffffff;
|
| 152 |
+
box-shadow: 0 0 10px rgba(59,130,246,0.5);
|
| 153 |
+
}
|
| 154 |
+
</style>
|
| 155 |
+
""", unsafe_allow_html=True)
|
| 156 |
+
|
| 157 |
+
# Initialize session state
|
| 158 |
+
if "user_query_input" not in st.session_state:
|
| 159 |
+
st.session_state["user_query_input"] = ""
|
| 160 |
+
if "show_more" not in st.session_state:
|
| 161 |
+
st.session_state["show_more"] = False
|
| 162 |
+
|
| 163 |
+
def set_user_query(q: str):
|
| 164 |
+
st.session_state["user_query_input"] = q
|
| 165 |
+
|
| 166 |
if doc_choice == "-- Select --":
|
| 167 |
st.info("β¬
οΈ Please choose a document from the sidebar to begin.")
|
| 168 |
else:
|
| 169 |
if doc_choice == "Sample PDF":
|
| 170 |
+
temp_path = os.path.join(os.path.dirname(__file__), "sample.pdf")
|
| 171 |
st.success("π Using built-in Sample PDF")
|
| 172 |
else:
|
| 173 |
uploaded_file = st.file_uploader("π Upload your PDF", type="pdf")
|
|
|
|
| 191 |
st.text_area("TOC Preview", toc_text, height=180)
|
| 192 |
query_suggestions = generate_dynamic_suggestions_from_toc(toc, chunks, os.path.basename(temp_path))
|
| 193 |
else:
|
| 194 |
+
query_suggestions = ["What is this document about?", "How do I start using this process?"]
|
|
|
|
|
|
|
|
|
|
| 195 |
|
| 196 |
with st.spinner("βοΈ Preparing search index..."):
|
| 197 |
embeddings = cache_embeddings(os.path.basename(temp_path), chunks, embed_chunks)
|
|
|
|
| 206 |
if query_suggestions:
|
| 207 |
st.markdown("#### π‘ Suggested Questions")
|
| 208 |
|
| 209 |
+
visible_suggestions = query_suggestions if st.session_state.show_more else query_suggestions[:3]
|
| 210 |
+
cols = st.columns(min(3, len(visible_suggestions)))
|
| 211 |
|
| 212 |
+
for i, q in enumerate(visible_suggestions):
|
| 213 |
+
cols[i % 3].button(f"π {q}", key=f"suggest_{i}", on_click=set_user_query, args=(q,))
|
|
|
|
|
|
|
|
|
|
| 214 |
|
| 215 |
toggle_text = "Show less β²" if st.session_state.show_more else "Show more βΌ"
|
| 216 |
+
if st.button(toggle_text):
|
| 217 |
st.session_state.show_more = not st.session_state.show_more
|
| 218 |
st.experimental_rerun()
|
| 219 |
|
| 220 |
+
user_query = st.text_input("Type your question or pick one above:", key="user_query_input")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 221 |
|
| 222 |
if user_query.strip():
|
| 223 |
st.caption("Mode: π§ Reasoning" if st.session_state.reasoning_mode else "Mode: π Strict Document")
|
|
|
|
| 224 |
with st.spinner("π Analyzing your document..."):
|
| 225 |
retrieved = retrieve_chunks(user_query, index, chunks, top_k=top_k, embeddings=embeddings)
|
| 226 |
answer = generate_answer(user_query, retrieved, reasoning_mode=st.session_state.reasoning_mode)
|
|
|
|
| 232 |
for i, r in enumerate(retrieved, start=1):
|
| 233 |
st.markdown(f"**Chunk {i}:** {r}")
|
| 234 |
|
|
|
|
|
|
|
|
|
|
| 235 |
if chunks:
|
| 236 |
st.markdown("---")
|
| 237 |
st.subheader("π Document Preview")
|