Hướng dẫn setup Ollama trên Google Colab kết nối VSCode miễn phí

import os
import subprocess
import time
import hashlib
from dotenv import load_dotenv

def setup():
    load_dotenv()
    token = os.getenv('NGROK_TOKEN')
    user = os.getenv('OLLAMA_USER', 'admin')
    raw_pass = os.getenv('OLLAMA_PASS', 'your-password-here')

    print("1. Installing Dependencies...")
    os.system("sudo apt-get update > /dev/null")
    os.system("sudo apt-get install zstd jq -y > /dev/null")

    print("2. Installing Ollama & Ngrok...")
    os.system("curl -fsSL https://ollama.com/install.sh | sh")
    os.system("curl -s https://ngrok-agent.s3.amazonaws.com/ngrok.asc | sudo tee /etc/apt/trusted.gpg.d/ngrok.asc >/dev/null")
    os.system('echo "deb https://ngrok-agent.s3.amazonaws.com buster main" | sudo tee /etc/apt/sources.list.d/ngrok.list > /dev/null')
    os.system("sudo apt update > /dev/null && sudo apt install ngrok -y > /dev/null")

    print("\n3. Starting Ollama Server...")
    os.environ['OLLAMA_HOST'] = '0.0.0.0'  # Allow external connections
    os.environ['OLLAMA_ORIGINS'] = '*'     # CORS for all origins
    os.system("pkill -9 ollama || true")
    os.system("nohup ollama serve > ollama.log 2>&1 &")

    # Wait for Ollama to start
    print("Waiting for Ollama to wake up...")
    for _ in range(10):
        if "127.0.0.1:11434" in subprocess.getoutput("netstat -tuln"):
            print("✅ Ollama is listening.")
            break
        time.sleep(2)

    print("\n4. Checking Model: gpt-oss:20b")
    models = subprocess.getoutput("ollama list")
    if "gpt-oss:20b" not in models:
        print("Model not found. Pulling now (this takes time)...")
        os.system("ollama pull gpt-oss:20b")  # ~12GB download
    else:
        print("✅ Model already exists.")

    print("\n5. Starting Ngrok Tunnel...")
    hashed_pass = hashlib.sha256(raw_pass.encode()).hexdigest()
    os.system("pkill -9 ngrok || true")
    os.system(f"nohup ngrok http 11434 --authtoken {token} --basic-auth {user}:{hashed_pass} > ngrok.log 2>&1 &")

    # Wait for tunnel URL
    print("Waiting for URL...")
    url = ""
    for _ in range(30):
        time.sleep(2)
        url = subprocess.getoutput("curl -s http://localhost:4040/api/tunnels | jq -r '.tunnels[0].public_url'")
        if url and url != "null": break

    if not url or url == "null":
        print("❌ Ngrok Error. Check ngrok.log")
        return

    print("\n" + "="*50)
    print(f"🚀 SERVER LIVE")
    print(f"MODEL: gpt-oss:20b")
    print(f"URL:   {url}")
    print("="*50)
    print(f'\n# Copy dòng này để chạy trên Windows PowerShell:')
    print(f'$env:OLLAMA_HOST="{url}"; $env:OLLAMA_USERNAME="{user}"; $env:OLLAMA_PASSWORD="{hashed_pass}"; python chat.py')
    print("="*50)

    # Keep alive
    while True:
        time.sleep(600)
        print(f"Heartbeat: {time.strftime('%H:%M:%S')}")

if __name__ == "__main__":
    setup()

from google.colab import drive
from google.colab import userdata
import os, sys

print("1. Mounting Drive...")
drive.mount('/content/drive')

path = '/content/drive/MyDrive/ollama-collab'
lib_path = os.path.join(path, 'lib')
model_path = os.path.join(path, 'ollama_models')

os.makedirs(lib_path, exist_ok=True)
os.makedirs(model_path, exist_ok=True)

print("2. Linking Models...")
!mkdir -p ~/.ollama
!ln -sfn {model_path} ~/.ollama/models  # Model sẽ persist trên Drive!

if lib_path not in sys.path:
    sys.path.insert(0, lib_path)

%cd {path}

print("3. Installing Requirements...")
with open('requirements.txt', 'w') as f: 
    f.write("python-dotenv \n requests")

if not os.path.exists(os.path.join(os.getcwd(), 'lib/requests')):
    !pip install --target={os.getcwd()}/lib -r requirements.txt
else:
    print("✅ Libraries found.")

print("4. Starting Server...")

# Get Secrets securely
try:
    ngrok_token = userdata.get('NGROK_TOKEN')
    ollama_pass = userdata.get('OLLAMA_PASS')
except:
    ngrok_token = "YOUR_TOKEN_HERE"
    ollama_pass = "YOUR_PASSWORD"
    print("⚠️ Secrets not found. Using fallback.")

os.environ['NGROK_TOKEN'] = ngrok_token
os.environ['OLLAMA_USER'] = 'admin'
os.environ['OLLAMA_PASS'] = ollama_pass

!python server_app.py

pip install flask requests

from flask import Flask, request, Response, stream_with_context
import requests
import os
import json

app = Flask(__name__)

# --- CONFIGURATION ---
# Thay bằng URL Ngrok từ Colab output
REMOTE_URL = os.getenv("OLLAMA_HOST", "https://your-ngrok-url.ngrok-free.app")
OLLAMA_USER = os.getenv("OLLAMA_USERNAME", "admin")
OLLAMA_PASS = os.getenv("OLLAMA_PASSWORD", "your-hashed-password-here")

# Headers để bypass ngrok browser warning
HEADERS = {
    "ngrok-skip-browser-warning": "true",
    "User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/120.0.0.0 Safari/537.36"
}
AUTH = (OLLAMA_USER, OLLAMA_PASS)

@app.route('/<path:path>', methods=['GET', 'POST', 'PUT', 'DELETE'])
def proxy(path):
    url = f"{REMOTE_URL}/{path}"
    method = request.method
    params = request.args

    print(f"DEBUG: Proxying {method} to {url}")

    try:
        # Parse và modify request payload
        payload = request.get_json(force=True, silent=True) or {}

        # LOGGING: Ghi log request để debug
        with open("vscode_requests.log", "a", encoding="utf-8") as log_file:
            log_file.write(f"\n--- REQUEST ({method}) ---\n")
            log_file.write(json.dumps(payload, indent=2))
            log_file.write("\n-------------------------\n")

        # FORCE MODEL: Luôn dùng gpt-oss:20b
        print(f"🛠️ Forcing Model: gpt-oss:20b")
        payload["model"] = "gpt-oss:20b"

        # CONTEXT INJECTION: Tối ưu memory
        if "options" not in payload:
            payload["options"] = {}
        payload["options"]["num_ctx"] = 2048      # Context window
        payload["options"]["num_predict"] = -1    # Unlimited generation
        payload["num_ctx"] = 8192                 # Fallback top-level

        # Forward request với authentication
        resp = requests.request(
            method=method,
            url=url,
            headers=HEADERS,
            json=payload,
            params=params,
            auth=AUTH,
            stream=True,
            timeout=300
        )

        # Loại bỏ hop-by-hop headers
        excluded_headers = ['content-encoding', 'content-length', 'transfer-encoding', 'connection']
        headers = [(name, value) for (name, value) in resp.raw.headers.items()
                   if name.lower() not in excluded_headers]

        return Response(
            stream_with_context(resp.iter_content(chunk_size=1024)), 
            status=resp.status_code, 
            headers=headers
        )

    except Exception as e:
        print(f"ERROR: {e}")
        return Response(json.dumps({"error": str(e)}), status=500)

if __name__ == '__main__':
    print("🚀 VS Code → Ollama Proxy Running")
    print(f"Targeting: {REMOTE_URL}")
    print("━" * 50)
    print("📝 Trong VSCode Continue, set API Base URL:")
    print("   http://localhost:11435")
    print("━" * 50)
    app.run(port=11435, threaded=True)

# Set environment variables (Lấy từ output của Colab)
$env:OLLAMA_HOST="https://your-ngrok-url.ngrok-free.app"
$env:OLLAMA_USERNAME="admin"
$env:OLLAMA_PASSWORD="your-hashed-password-from-colab"

# Chạy proxy
python vscode_proxy.py

  - name: Remote Ollama (20B)
    provider: ollama
    model: gpt-oss:20b
    apiBase: http://localhost:11435
    roles:
      - chat
      - edit
      - apply