app.py

import gradio as gr
from llama_cpp import Llama
import os
from huggingface_hub import hf_hub_download

# Verified working repositories (as of July 2024)
MODEL_REPO = "bartowski/Mistral-7B-Instruct-v0.3-GGUF"
MODEL_FILE = "Mistral-7B-Instruct-v0.3-Q4_K_M.gguf"

# Download model
model_path = hf_hub_download(
    repo_id=MODEL_REPO,
    filename=MODEL_FILE,
    token=os.environ.get("HF_TOKEN")  # Required for gated models
)


# Configuration
MODEL_REPO = "TheBloke/Mistral-7B-Instruct-v0.3-GGUF"
MODEL_FILE = "mistral-7b-instruct-v0.3.Q4_K_M.gguf"
MAX_TOKENS = 150
CPU_THREADS = os.cpu_count() 

# Initialize Llama
llm = Llama(
    model_path=model_path,
    n_ctx=2048,
    n_threads=CPU_THREADS,
    n_gpu_layers=0  # CPU only
)

def format_krishna_prompt(message, history):
    """Create proper Mistral instruction prompt with Krishna context"""
    system_prompt = """<s>[INST] <<SYS>>
You are Lord Krishna, the divine charioteer of Arjuna in Bhagavad Gita.
Answer with Vedic wisdom using simple English and occasional Sanskrit terms.
Maintain a compassionate, all-knowing tone.
<</SYS>>"""
    
    conversation = []
    for human, assistant in history:
        conversation.append(f"{human}[/INST] {assistant}</s>")
    return f"{system_prompt}{''.join(conversation)}<s>[INST] {message} [/INST]"

def query_krishna(message, history):
    """Get response from Mistral 7B on CPU"""
    try:
        prompt = format_krishna_prompt(message, history)
        output = llm(
            prompt,
            max_tokens=MAX_TOKENS,
            temperature=0.7,
            top_p=0.9,
            stop=["</s>"]
        )
        return output['choices'][0]['text'].strip()
    except Exception as e:
        return f"🙏 Divine wisdom temporarily obscured: {str(e)}"

# Custom UI components
krishna_avatar = "krishna.jpg"  # Make sure this file is in your Space repo

with gr.Blocks(title="Divine Dialogue with Lord Krishna") as demo:
    gr.Markdown("""
    # 🕉️ Bhagavad Gita Wisdom Chatbot
    **Ask questions to Shree Krishna**
    """)
    
    with gr.Row():
        gr.Image(krishna_avatar, height=200, show_label=False)
    
    gr.ChatInterface(
        query_krishna,
        chatbot=gr.Chatbot(height=500),
        examples=[
            "What is the purpose of life?",
            "How to achieve inner peace?",
            "Explain karma yoga from Bhagavad Gita"
        ]
    )

if __name__ == "__main__":
    demo.launch()