Update app.py

app.py

@@ -1,5 +1,7 @@
+!pip install langdetect faiss-cpu transformers gradio groq sentence-transformers pypdf2 python-pptx pandas docx2txt
+
 import gradio as gr
-import fitz
+import fitz  # PyMuPDF
 import numpy as np
 import requests
 import faiss
@@ -12,16 +14,18 @@ from sentence_transformers import SentenceTransformer
 from concurrent.futures import ThreadPoolExecutor
 
 # Configuration
-GROQ_API_KEY = "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"    # 🔑 REPLACE WITH YOUR ACTUAL KEY
-MODEL_NAME = "groq-llama3"
+GROQ_API_KEY = "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"  # Replace with your actual key
+EMBEDDING_MODEL = "sentence-transformers/all-MiniLM-L6-v2"  # Proper embedding model
 CHUNK_SIZE = 512
 MAX_TOKENS = 4096
-MODEL = SentenceTransformer(MODEL_NAME)
 WORKERS = 8
 
+# Initialize the embedding model
+embedding_model = SentenceTransformer(EMBEDDING_MODEL)
+
 class DocumentProcessor:
     def __init__(self):
-        self.index = faiss.IndexFlatIP(MODEL.get_sentence_embedding_dimension())
+        self.index = faiss.IndexFlatIP(embedding_model.get_sentence_embedding_dimension())
         self.chunks = []
         self.processor_pool = ThreadPoolExecutor(max_workers=WORKERS)
 
@@ -79,9 +83,22 @@ class DocumentProcessor:
             return ""
 
     def semantic_chunking(self, text):
-        words = re.findall(r'\S+\s*', text)
-        chunks = [''.join(words[i:i+CHUNK_SIZE//2]) for i in range(0, len(words), CHUNK_SIZE//2)]
-        return chunks[:1000]
+        sentences = re.split(r'(?<!\w\.\w.)(?<![A-Z][a-z]\.)(?<=\.|\?)\s', text)
+        chunks = []
+        current_chunk = ""
+        
+        for sentence in sentences:
+            if len(current_chunk) + len(sentence) < CHUNK_SIZE:
+                current_chunk += " " + sentence
+            else:
+                if current_chunk:
+                    chunks.append(current_chunk.strip())
+                current_chunk = sentence
+        
+        if current_chunk:
+            chunks.append(current_chunk.strip())
+        
+        return chunks[:1000]  # Limit to 1000 chunks per document
 
     def process_documents(self, files):
         self.chunks = []
@@ -101,9 +118,9 @@ class DocumentProcessor:
             return "Error: No chunks generated from documents"
 
         try:
-            embeddings = MODEL.encode(
+            embeddings = embedding_model.encode(
                 all_chunks,
-                batch_size=512,
+                batch_size=32,
                 convert_to_tensor=True,
                 show_progress_bar=False
             ).cpu().numpy().astype('float32')
@@ -124,7 +141,7 @@ class DocumentProcessor:
             print("\n" + "="*40 + " QUERY PROCESSING " + "="*40)
             print(f"Question: {question}")
 
-            question_embedding = MODEL.encode([question], convert_to_tensor=True).cpu().numpy().astype('float32')
+            question_embedding = embedding_model.encode([question], convert_to_tensor=True).cpu().numpy().astype('float32')
             _, indices = self.index.search(question_embedding, 3)
             print(f"Top indices: {indices}")
 
@@ -139,12 +156,12 @@ class DocumentProcessor:
             payload = {
                 "messages": [{
                     "role": "user",
-                    "content": f"Answer concisely: {question}\nContext: {context}"
+                    "content": f"Answer concisely based on the context: {question}\nContext: {context}"
                 }],
                 "model": "mixtral-8x7b-32768",
                 "temperature": 0.3,
                 "max_tokens": MAX_TOKENS,
-                "stream": True
+                "stream": False  # Changed to False for simpler handling
             }
 
             response = requests.post(
@@ -159,20 +176,8 @@ class DocumentProcessor:
             if response.status_code != 200:
                 return f"API Error: {response.text}", False
 
-            full_answer = []
-            for chunk in response.iter_lines():
-                if chunk:
-                    try:
-                        decoded = chunk.decode('utf-8').strip()
-                        if decoded.startswith('data:'):
-                            data = json.loads(decoded[5:])
-                            if content := data.get('choices', [{}])[0].get('delta', {}).get('content', ''):
-                                full_answer.append(content)
-                    except Exception as e:
-                        print(f"Chunk Error: {str(e)}")
-                        continue
-
-            final_answer = ''.join(full_answer)
+            data = response.json()
+            final_answer = data.get("choices", [{}])[0].get("message", {}).get("content", "")
             print(f"Final Answer: {final_answer}")
             return final_answer, True
 
@@ -189,19 +194,23 @@ def ask_question(question, chat_history):
     answer, success = processor.query(question)
     return chat_history + [(question, answer)]
 
-with gr.Blocks(title="System") as app:
-    gr.Markdown("## 🚀 Multi-Format-Reader ChatBot")
+with gr.Blocks(title="Document ChatBot") as app:
+    gr.Markdown("## 🚀 Multi-Format Document ChatBot")
     with gr.Row():
-        files = gr.File(file_count="multiple",
-                      file_types=[".pdf", ".docx", ".txt", ".pptx", ".xls", ".xlsx", ".csv"],
-                      label="Upload Documents")
-        process_btn = gr.Button("Process", variant="primary")
+        files = gr.File(
+            file_count="multiple",
+            file_types=[".pdf", ".docx", ".txt", ".pptx", ".xls", ".xlsx", ".csv"],
+            label="Upload Documents"
+        )
+        process_btn = gr.Button("Process Documents", variant="primary")
     status = gr.Textbox(label="Processing Status", interactive=False)
     chatbot = gr.Chatbot(height=500, label="Chat History")
     with gr.Row():
-        question = gr.Textbox(label="Your Query",
-                            placeholder="Enter your question...",
-                            max_lines=3)
+        question = gr.Textbox(
+            label="Your Query",
+            placeholder="Enter your question about the documents...",
+            max_lines=3
+        )
         ask_btn = gr.Button("Ask", variant="primary")
     clear_btn = gr.Button("Clear Chat")