Update VED-app.py

VED-app.py

@@ -0,0 +1,133 @@
+app_code = '''
+import streamlit as st
+from transformers import AutoTokenizer, AutoModelForCausalLM
+import chromadb
+from sentence_transformers import SentenceTransformer
+import torch
+import time
+
+st.set_page_config(
+    page_title="VED — India's AI",
+    page_icon="🇮🇳",
+    layout="centered"
+)
+
+st.title("VED 🇮🇳")
+st.caption("India's Own AI — Built by PRANTH1304")
+st.divider()
+
+@st.cache_resource
+def load_everything():
+    # Load embedder
+    embedder = SentenceTransformer("all-MiniLM-L6-v2")
+    
+    # Load knowledge base
+    client = chromadb.Client()
+    collection = client.create_collection("ved_knowledge")
+    
+    knowledge = [
+        ("startup_001", "To register a startup in India, visit startupindia.gov.in, get DPIIT recognition, and enjoy 3 years tax exemption under Section 80-IAC."),
+        ("startup_002", "GST registration is mandatory in India if annual turnover exceeds 20 lakhs. It gives input tax credit which reduces overall tax burden."),
+        ("startup_003", "Y Combinator gives 500000 dollars for 7 percent equity. Apply at ycombinator.com with a working MVP and real users."),
+        ("startup_004", "Startup India Seed Fund gives up to 20 lakhs free money to early stage Indian startups. No equity taken. Apply at startupindia.gov.in."),
+        ("newton_001", "Newton first law states that an object stays at rest or moves at constant velocity unless an external force acts on it."),
+        ("tajmahal_001", "The Taj Mahal was built by Mughal Emperor Shah Jahan between 1632 and 1653 in memory of his wife Mumtaz Mahal in Agra."),
+        ("bigo_001", "Big O notation measures algorithm efficiency. O(1) is constant time. O(n) is linear. O(log n) is logarithmic. O(n squared) is quadratic worst case."),
+        ("binary_001", "Binary search works by comparing the middle element of a sorted array with the target. Search left half if smaller, right half if larger. Time complexity O(log n)."),
+        ("photo_001", "Photosynthesis is the process where plants use sunlight, water, and carbon dioxide to produce food. Chlorophyll absorbs sunlight. Oxygen is released as byproduct."),
+        ("solid_001", "SOLID principles: Single responsibility, Open closed, Liskov substitution, Interface segregation, Dependency inversion. These make code clean and maintainable."),
+        ("gandhi_001", "Mahatma Gandhi led India independence through non-violence. Key movements: Non-Cooperation 1920, Salt March 1930, Quit India 1942."),
+        ("rag_001", "RAG means Retrieval Augmented Generation. Documents stored as embeddings. Relevant chunks retrieved for each query. Model answers using retrieved context."),
+        ("recursion_001", "Recursion is when a function calls itself. Every recursive function needs a base case to stop. Example: factorial of n equals n times factorial of n minus 1."),
+        ("india_001", "India is the world largest democracy with 1.4 billion people. Parliamentary system with Lok Sabha and Rajya Sabha. Prime Minister is head of government."),
+        ("python_001", "Python list comprehension: [x for x in range(10) if x percent 2 == 0] gives even numbers. Faster and cleaner than for loops."),
+    ]
+    
+    texts      = [item[1] for item in knowledge]
+    ids        = [item[0] for item in knowledge]
+    embeddings = embedder.encode(texts).tolist()
+    collection.add(documents=texts, embeddings=embeddings, ids=ids)
+    
+    # Load model
+    tokenizer = AutoTokenizer.from_pretrained("ved_mistral")
+    model     = AutoModelForCausalLM.from_pretrained(
+        "ved_mistral",
+        torch_dtype=torch.float16,
+        device_map="auto"
+    )
+    
+    return embedder, collection, tokenizer, model
+
+embedder, collection, tokenizer, model = load_everything()
+
+def ask_ved(question):
+    query_emb = embedder.encode([question]).tolist()
+    results   = collection.query(query_embeddings=query_emb, n_results=1)
+    context   = results["documents"][0][0]
+
+    prompt = f"""[INST] You are VED, India AI assistant.
+Use ONLY the context below. One complete sentence answer only.
+
+Context: {context}
+Question: {question} [/INST]"""
+
+    inputs       = tokenizer(prompt, return_tensors="pt").to("cuda")
+    input_length = inputs["input_ids"].shape[1]
+
+    with torch.no_grad():
+        outputs = model.generate(
+            **inputs,
+            max_new_tokens=100,
+            temperature=0.1,
+            do_sample=False,
+            repetition_penalty=1.3,
+            pad_token_id=tokenizer.eos_token_id
+        )
+
+    response = tokenizer.decode(
+        outputs[0][input_length:],
+        skip_special_tokens=True
+    ).strip()
+
+    if "." in response:
+        response = response[:response.index(".")+1]
+
+    return response.strip()
+
+# Chat interface
+if "messages" not in st.session_state:
+    st.session_state.messages = []
+    st.session_state.messages.append({
+        "role": "assistant",
+        "content": "Namaste! I am VED — India's AI. Ask me anything about startups, science, history, coding, or India. 🇮🇳"
+    })
+
+for message in st.session_state.messages:
+    with st.chat_message(message["role"]):
+        st.write(message["content"])
+
+if question := st.chat_input("Ask VED anything..."):
+    st.session_state.messages.append({"role": "user", "content": question})
+    with st.chat_message("user"):
+        st.write(question)
+
+    with st.chat_message("assistant"):
+        with st.spinner("VED is thinking..."):
+            answer = ask_ved(question)
+        st.write(answer)
+        st.session_state.messages.append({"role": "assistant", "content": answer})
+
+st.divider()
+col1, col2, col3 = st.columns(3)
+col1.metric("Model", "Mistral 7B")
+col2.metric("Knowledge", "15 chunks")
+col3.metric("Built by", "PRANTH1304")
+'''
+
+with open("app.py", "w") as f:
+    f.write(app_code)
+
+print("app.py created!")
+print("Now go to HuggingFace Space: PRANTH1304/ved-app")
+print("Replace app.py with this new code")
+print("VED will be a full chat interface!")