Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -5,9 +5,11 @@ from groq import Groq
|
|
| 5 |
from sentence_transformers import SentenceTransformer
|
| 6 |
from pypdf import PdfReader
|
| 7 |
import gradio as gr
|
|
|
|
|
|
|
|
|
|
| 8 |
|
| 9 |
# -------------------- Configuration --------------------
|
| 10 |
-
# Get API key from environment variable (set as a secret on Hugging Face)
|
| 11 |
GROQ_API_KEY = os.getenv("GROQ_API_KEY")
|
| 12 |
if not GROQ_API_KEY:
|
| 13 |
raise ValueError("GROQ_API_KEY environment variable not set.")
|
|
@@ -17,7 +19,6 @@ embedding_model = SentenceTransformer("all-MiniLM-L6-v2")
|
|
| 17 |
|
| 18 |
# -------------------- PDF processing --------------------
|
| 19 |
def extract_text_from_pdf(pdf_file):
|
| 20 |
-
"""Extract text from a PDF file."""
|
| 21 |
reader = PdfReader(pdf_file)
|
| 22 |
text = ""
|
| 23 |
for page in reader.pages:
|
|
@@ -27,18 +28,15 @@ def extract_text_from_pdf(pdf_file):
|
|
| 27 |
return text.strip()
|
| 28 |
|
| 29 |
def chunk_text(text, chunk_size=800, overlap=150):
|
| 30 |
-
"""Split text into overlapping chunks."""
|
| 31 |
chunks = []
|
| 32 |
start = 0
|
| 33 |
while start < len(text):
|
| 34 |
end = start + chunk_size
|
| 35 |
-
|
| 36 |
-
chunks.append(chunk)
|
| 37 |
start = end - overlap
|
| 38 |
return chunks
|
| 39 |
|
| 40 |
def create_faiss_index(chunks):
|
| 41 |
-
"""Build a FAISS index from text chunks."""
|
| 42 |
embeddings = embedding_model.encode(chunks)
|
| 43 |
embeddings = np.array(embeddings).astype("float32")
|
| 44 |
dimension = embeddings.shape[1]
|
|
@@ -47,7 +45,6 @@ def create_faiss_index(chunks):
|
|
| 47 |
return index
|
| 48 |
|
| 49 |
def retrieve(query, index, chunks, top_k=3):
|
| 50 |
-
"""Retrieve most relevant chunks for a query."""
|
| 51 |
query_embedding = embedding_model.encode([query])
|
| 52 |
query_embedding = np.array(query_embedding).astype("float32")
|
| 53 |
distances, indices = index.search(query_embedding, top_k)
|
|
@@ -58,7 +55,6 @@ def retrieve(query, index, chunks, top_k=3):
|
|
| 58 |
return retrieved
|
| 59 |
|
| 60 |
def ask_llama(context, question):
|
| 61 |
-
"""Query Groq's Llama model with context."""
|
| 62 |
prompt = f"""
|
| 63 |
You are a pharmaceutical expert assistant.
|
| 64 |
|
|
@@ -81,12 +77,11 @@ Question:
|
|
| 81 |
except Exception as e:
|
| 82 |
return f"Groq API Error: {str(e)}"
|
| 83 |
|
| 84 |
-
# -------------------- Global state
|
| 85 |
index = None
|
| 86 |
chunks = None
|
| 87 |
|
| 88 |
def process_pdf(pdf):
|
| 89 |
-
"""Upload and index a PDF."""
|
| 90 |
global index, chunks
|
| 91 |
if pdf is None:
|
| 92 |
return "⚠️ Please upload a PDF."
|
|
@@ -98,7 +93,6 @@ def process_pdf(pdf):
|
|
| 98 |
return "✅ Document processed successfully!"
|
| 99 |
|
| 100 |
def chatbot(question):
|
| 101 |
-
"""Answer a question using the indexed document."""
|
| 102 |
global index, chunks
|
| 103 |
if index is None:
|
| 104 |
return "⚠️ Please upload and process a PDF first."
|
|
@@ -136,7 +130,7 @@ textarea {
|
|
| 136 |
}
|
| 137 |
"""
|
| 138 |
|
| 139 |
-
with gr.Blocks(
|
| 140 |
gr.Markdown("""
|
| 141 |
<h1 style='text-align:center; font-size:48px; font-weight:800;
|
| 142 |
background: linear-gradient(90deg,#ff6a00,#ee0979,#667eea);
|
|
@@ -165,11 +159,9 @@ with gr.Blocks(css=custom_css) as demo:
|
|
| 165 |
label="🧠 AI Answer",
|
| 166 |
lines=18,
|
| 167 |
max_lines=30,
|
| 168 |
-
show_copy_button=True,
|
| 169 |
interactive=False
|
| 170 |
)
|
| 171 |
ask_button.click(fn=chatbot, inputs=question_input, outputs=answer_output)
|
| 172 |
|
| 173 |
-
# Launch
|
| 174 |
-
|
| 175 |
-
demo.launch()
|
|
|
|
| 5 |
from sentence_transformers import SentenceTransformer
|
| 6 |
from pypdf import PdfReader
|
| 7 |
import gradio as gr
|
| 8 |
+
import nest_asyncio
|
| 9 |
+
|
| 10 |
+
nest_asyncio.apply()
|
| 11 |
|
| 12 |
# -------------------- Configuration --------------------
|
|
|
|
| 13 |
GROQ_API_KEY = os.getenv("GROQ_API_KEY")
|
| 14 |
if not GROQ_API_KEY:
|
| 15 |
raise ValueError("GROQ_API_KEY environment variable not set.")
|
|
|
|
| 19 |
|
| 20 |
# -------------------- PDF processing --------------------
|
| 21 |
def extract_text_from_pdf(pdf_file):
|
|
|
|
| 22 |
reader = PdfReader(pdf_file)
|
| 23 |
text = ""
|
| 24 |
for page in reader.pages:
|
|
|
|
| 28 |
return text.strip()
|
| 29 |
|
| 30 |
def chunk_text(text, chunk_size=800, overlap=150):
|
|
|
|
| 31 |
chunks = []
|
| 32 |
start = 0
|
| 33 |
while start < len(text):
|
| 34 |
end = start + chunk_size
|
| 35 |
+
chunks.append(text[start:end])
|
|
|
|
| 36 |
start = end - overlap
|
| 37 |
return chunks
|
| 38 |
|
| 39 |
def create_faiss_index(chunks):
|
|
|
|
| 40 |
embeddings = embedding_model.encode(chunks)
|
| 41 |
embeddings = np.array(embeddings).astype("float32")
|
| 42 |
dimension = embeddings.shape[1]
|
|
|
|
| 45 |
return index
|
| 46 |
|
| 47 |
def retrieve(query, index, chunks, top_k=3):
|
|
|
|
| 48 |
query_embedding = embedding_model.encode([query])
|
| 49 |
query_embedding = np.array(query_embedding).astype("float32")
|
| 50 |
distances, indices = index.search(query_embedding, top_k)
|
|
|
|
| 55 |
return retrieved
|
| 56 |
|
| 57 |
def ask_llama(context, question):
|
|
|
|
| 58 |
prompt = f"""
|
| 59 |
You are a pharmaceutical expert assistant.
|
| 60 |
|
|
|
|
| 77 |
except Exception as e:
|
| 78 |
return f"Groq API Error: {str(e)}"
|
| 79 |
|
| 80 |
+
# -------------------- Global state --------------------
|
| 81 |
index = None
|
| 82 |
chunks = None
|
| 83 |
|
| 84 |
def process_pdf(pdf):
|
|
|
|
| 85 |
global index, chunks
|
| 86 |
if pdf is None:
|
| 87 |
return "⚠️ Please upload a PDF."
|
|
|
|
| 93 |
return "✅ Document processed successfully!"
|
| 94 |
|
| 95 |
def chatbot(question):
|
|
|
|
| 96 |
global index, chunks
|
| 97 |
if index is None:
|
| 98 |
return "⚠️ Please upload and process a PDF first."
|
|
|
|
| 130 |
}
|
| 131 |
"""
|
| 132 |
|
| 133 |
+
with gr.Blocks() as demo:
|
| 134 |
gr.Markdown("""
|
| 135 |
<h1 style='text-align:center; font-size:48px; font-weight:800;
|
| 136 |
background: linear-gradient(90deg,#ff6a00,#ee0979,#667eea);
|
|
|
|
| 159 |
label="🧠 AI Answer",
|
| 160 |
lines=18,
|
| 161 |
max_lines=30,
|
|
|
|
| 162 |
interactive=False
|
| 163 |
)
|
| 164 |
ask_button.click(fn=chatbot, inputs=question_input, outputs=answer_output)
|
| 165 |
|
| 166 |
+
# Launch with custom CSS (Gradio 6+ compatible)
|
| 167 |
+
demo.launch(css=custom_css)
|
|
|