app.py

import gradio as gr
import subprocess
import os
import threading
import sys
import time
from pathlib import Path

# Training log file
TRAINING_LOG = "/tmp/training.log"
TRAINING_PID_FILE = "/tmp/training.pid"

def start_training():
    """Start D1337 CIPHER training - SUBPROCESS VERSION (MORE RELIABLE)"""
    
    # Check if already running
    if Path(TRAINING_PID_FILE).exists():
        try:
            pid = int(Path(TRAINING_PID_FILE).read_text())
            # Check if process still running
            os.kill(pid, 0)  # Will raise if not running
            return "⚠️ Training already running! Check logs below..."
        except:
            # Process dead, remove PID file
            Path(TRAINING_PID_FILE).unlink()
    
    # Clear log file
    Path(TRAINING_LOG).write_text("🔥 D1337 CIPHER TRAINING STARTING...\n\n", encoding="utf-8")
    
    # Use optimized training script for L40S x4
    runner_path = "/app/train.py"
    
    # Start subprocess
    def run_training():
        try:
            # Run training script and redirect output to log
            with open(TRAINING_LOG, "a", encoding="utf-8") as log_file:
                process = subprocess.Popen(
                    [sys.executable, runner_path],
                    stdout=log_file,
                    stderr=subprocess.STDOUT,
                    cwd="/app",
                    env=os.environ.copy()
                )
                
                # Write PID
                Path(TRAINING_PID_FILE).write_text(str(process.pid))
                
                # Wait for completion
                process.wait()
                
                # Remove PID file
                if Path(TRAINING_PID_FILE).exists():
                    Path(TRAINING_PID_FILE).unlink()
                    
        except Exception as e:
            error_msg = f"\n❌ ERROR: {str(e)}\n"
            Path(TRAINING_LOG).write_text(error_msg, encoding="utf-8", mode="a")
            if Path(TRAINING_PID_FILE).exists():
                Path(TRAINING_PID_FILE).unlink()
    
    # Run in background thread
    thread = threading.Thread(target=run_training)
    thread.daemon = False
    thread.start()
    
    return "🔥 D1337 CIPHER TRAINING STARTED!\n\nLoading model (31B) - this may take 2-5 minutes...\n\nOutput will appear below automatically (refresh every 3 seconds)."

def get_training_log():
    """Get latest training log"""
    try:
        if Path(TRAINING_LOG).exists():
            content = Path(TRAINING_LOG).read_text(encoding="utf-8")
            if content.strip():
                return content
        return "Waiting for training to start...\n\nClick 'START TRAINING' button above."
    except Exception as e:
        return f"Error reading log: {str(e)}"

# UI
with gr.Blocks(title="D1337 CIPHER Training") as demo:
    gr.Markdown("# 🔥 D1337 CIPHER C2 V.1 - TRAINING")
    gr.Markdown("**Hardware**: L40S x4 (192GB VRAM)")
    gr.Markdown("**Base**: Huihui-GLM-4.7-Flash-abliterated (31B)")
    gr.Markdown("**Dataset**: 92 samples | **Epochs**: 3 | **4-bit + LoRA**")
    
    with gr.Row():
        train_btn = gr.Button("🚀 START TRAINING", variant="primary")
    
    output = gr.Textbox(label="Training Output", lines=20, value="Click 'START TRAINING' to begin...")
    
    refresh_btn = gr.Button("🔄 Refresh Logs", variant="secondary")
    
    train_btn.click(start_training, outputs=output)
    refresh_btn.click(get_training_log, outputs=output)
    
    # Auto-load on page refresh
    demo.load(fn=get_training_log, outputs=output)
    
    gr.Markdown("**Expected time: ~5-10 minutes on L40S x4**")

demo.launch(server_name="0.0.0.0", server_port=7860, share=False, ssl_verify=False, show_error=True)