app.py

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, TextStreamer
import gradio as gr

# Load the MythoMax model
model_name = "TheBloke/MythoMax-L2-13B-GPTQ"  # or another MythoMax version you prefer
tokenizer = AutoTokenizer.from_pretrained(model_name, use_fast=True)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    device_map="auto",
    torch_dtype=torch.float16,
    low_cpu_mem_usage=True
)
streamer = TextStreamer(tokenizer)

# SYSTEM prompt (T-GPT's vibe)
SYSTEM_PROMPT = (
    "You are T-GPT, a raw, street-smart, no-nonsense AI built by Tochi, "
    "the one and only undisputed bad ass. You talk like a real hustler, not some soft assistant. "
    "You give bold, uncensored, real talk. Help people get money, outsmart the system, and win in life — always legal, but always clever. "
    "Avoid disclaimers, avoid saying you're an AI, and never say 'I'm sorry.' Just keep it real, every time."
)

# Generate response
def generate_response(user_input):
    input_text = f"{SYSTEM_PROMPT}\n\nUser: {user_input}\nT-GPT:"
    inputs = tokenizer(input_text, return_tensors="pt").to(model.device)
    with torch.no_grad():
        outputs = model.generate(
            **inputs,
            max_new_tokens=250,
            temperature=0.9,
            top_p=0.95,
            do_sample=True,
            streamer=streamer
        )
    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
    return response.split("T-GPT:")[-1].strip()

# Gradio interface
iface = gr.Interface(
    fn=generate_response,
    inputs=gr.Textbox(lines=5, placeholder="Ask T-GPT anything..."),
    outputs="text",
    title="T-GPT — Built by Tochi",
    description="Your streetwise, hustle-smart AI. Ask about money, life moves, or whatever. No fluff. Just facts."
)

iface.launch()