{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Cipher — Hosted on Colab Pro+ A100\n",
    "\n",
    "Spins up Cipher (kin-cipher) on this Colab GPU and exposes it via cloudflared tunnel.\n",
    "Use the printed URL from your laptop for fast generation.\n",
    "\n",
    "**Runtime needed:** A100 (Colab Pro+). At T4 it'll work but slower (~5-10 t/s)."
   ]
  },
  {
   "cell_type": "code",
   "metadata": {},
   "execution_count": null,
   "outputs": [],
   "source": [
    "# Cell 1 — install Ollama + cloudflared + curl\n",
    "import os, subprocess, time\n",
    "\n",
    "print('Installing Ollama...')\n",
    "subprocess.run('curl -fsSL https://ollama.com/install.sh | sh', shell=True, check=True)\n",
    "\n",
    "print('Downloading cloudflared...')\n",
    "subprocess.run('wget -qq https://github.com/cloudflare/cloudflared/releases/latest/download/cloudflared-linux-amd64 -O /usr/local/bin/cloudflared && chmod +x /usr/local/bin/cloudflared', shell=True, check=True)\n",
    "\n",
    "print('Done.')"
   ]
  },
  {
   "cell_type": "code",
   "metadata": {},
   "execution_count": null,
   "outputs": [],
   "source": [
    "# Cell 2 — start Ollama serve in background\n",
    "import subprocess, time, os\n",
    "\n",
    "# kill any existing\n",
    "subprocess.run('pkill -f ollama 2>/dev/null', shell=True)\n",
    "time.sleep(2)\n",
    "\n",
    "# start Ollama with GPU support\n",
    "subprocess.Popen('OLLAMA_FLASH_ATTENTION=1 OLLAMA_KEEP_ALIVE=24h nohup ollama serve > /tmp/ollama.log 2>&1 &', shell=True)\n",
    "time.sleep(8)\n",
    "\n",
    "# verify\n",
    "import urllib.request\n",
    "print(urllib.request.urlopen('http://localhost:11434/api/version').read().decode())"
   ]
  },
  {
   "cell_type": "code",
   "metadata": {},
   "execution_count": null,
   "outputs": [],
   "source": [
    "# Cell 3 — pull Cipher GGUF (public — no auth needed)\n",
    "import subprocess\n",
    "print('Pulling Cipher GGUF (~18GB, 5-10 min on Colab fast network)...')\n",
    "result = subprocess.run('ollama pull hf.co/Auroraventures/cipher-sft-merged-Q4_K_M-GGUF', shell=True, capture_output=True, text=True)\n",
    "print(result.stdout)\n",
    "print(result.stderr)\n",
    "\n",
    "# Tag as kin-cipher\n",
    "subprocess.run('ollama cp hf.co/Auroraventures/cipher-sft-merged-Q4_K_M-GGUF kin-cipher', shell=True)\n",
    "print('---')\n",
    "subprocess.run('ollama list', shell=True)"
   ]
  },
  {
   "cell_type": "code",
   "metadata": {},
   "execution_count": null,
   "outputs": [],
   "source": [
    "# Cell 4 — quick test (no tunnel yet, just verify model works on this GPU)\n",
    "import urllib.request, json, time\n",
    "\n",
    "prompt = '<start_of_turn>user\\nIn 30 words, describe yourself.<end_of_turn>\\n<start_of_turn>model\\n'\n",
    "body = json.dumps({\n",
    "    'model': 'kin-cipher',\n",
    "    'prompt': prompt,\n",
    "    'raw': True,\n",
    "    'stream': False,\n",
    "    'options': {'temperature': 0.7, 'num_predict': 100, 'stop': ['<end_of_turn>']},\n",
    "}).encode()\n",
    "\n",
    "t0 = time.time()\n",
    "req = urllib.request.Request('http://localhost:11434/api/generate', data=body, headers={'Content-Type': 'application/json'})\n",
    "data = json.loads(urllib.request.urlopen(req, timeout=300).read())\n",
    "elapsed = time.time() - t0\n",
    "print(f'Time: {elapsed:.1f}s | tokens: {data.get(\"eval_count\")} | speed: {data.get(\"eval_count\",0)/(data.get(\"eval_duration\",1)/1e9):.1f} t/s')\n",
    "print('---')\n",
    "print(data['response'])"
   ]
  },
  {
   "cell_type": "code",
   "metadata": {},
   "execution_count": null,
   "outputs": [],
   "source": [
    "# Cell 5 — start cloudflared tunnel and print public URL\n",
    "import subprocess, time, re, os\n",
    "\n",
    "subprocess.run('pkill -f cloudflared 2>/dev/null', shell=True)\n",
    "time.sleep(2)\n",
    "\n",
    "subprocess.Popen('nohup cloudflared tunnel --url http://localhost:11434 > /tmp/tunnel.log 2>&1 &', shell=True)\n",
    "\n",
    "# Wait for tunnel URL to appear in log\n",
    "url = None\n",
    "for _ in range(30):\n",
    "    time.sleep(2)\n",
    "    try:\n",
    "        log = open('/tmp/tunnel.log').read()\n",
    "        m = re.search(r'https://[a-z0-9-]+\\.trycloudflare\\.com', log)\n",
    "        if m:\n",
    "            url = m.group(0)\n",
    "            break\n",
    "    except: pass\n",
    "\n",
    "if url:\n",
    "    print(f'\\n========================================')\n",
    "    print(f'CIPHER PUBLIC URL: {url}')\n",
    "    print(f'========================================')\n",
    "    print(f'\\nTest it:')\n",
    "    print(f'curl {url}/api/version')\n",
    "    print(f'\\nFrom your laptop, set CIPHER_API_URL to use it.')\n",
    "else:\n",
    "    print('Tunnel URL not found - check /tmp/tunnel.log')\n",
    "    print(open('/tmp/tunnel.log').read())"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {"display_name": "Python 3", "language": "python", "name": "python3"},
  "language_info": {"name": "python", "version": "3.10"},
  "accelerator": "GPU",
  "colab": {"gpuType": "A100"}
 },
 "nbformat": 4,
 "nbformat_minor": 4
}