使用 Colab CLI 按需部署 Qwen3.8-27B GGUF

!pip install -q -U "huggingface_hub[hf_transfer]"

from google.colab import drive
drive.mount('/content/drive')

from huggingface_hub import hf_hub_download

REPO = "JonathanColetti/Qwen3.8-27B-Uncensored-GGUF"
FILE = "Qwen3.8-27B-Uncensored-Q8_0.gguf"
LOCAL = "/content/qwen38_download"
DRIVE_DIR = "/content/drive/MyDrive/colab/Qwen3.8-27B-Uncensored-GGUF"

model_path = hf_hub_download(
    repo_id=REPO,
    filename=FILE,
    local_dir=LOCAL,
)

!mkdir -p "{DRIVE_DIR}"
!rsync -ah --partial --progress \
    "{model_path}" "{DRIVE_DIR}/"

from google.colab import drive
drive.mount('/content/drive')

import secrets
import subprocess
import time
import urllib.request

SERVER = '/content/llama-server'
MODEL = '/content/models/Qwen3.8-27B-Uncensored-Q8_0.gguf'
API_KEY = secrets.token_urlsafe(32)

command = [
    SERVER,
    '-m', MODEL,
    '--alias', 'qwen3.8-27b',
    '--api-key', API_KEY,
    '-ngl', '99',
    '-c', '65536',
    '-ctk', 'q8_0',
    '-ctv', 'q8_0',
    '-np', '1',
    '--jinja',
    '--host', '127.0.0.1',
    '--port', '8000',
]

log = open('/content/llama-server.log', 'w')
subprocess.Popen(command, stdout=log, stderr=subprocess.STDOUT)

for _ in range(90):
    try:
        if urllib.request.urlopen(
            'http://127.0.0.1:8000/health', timeout=3
        ).status == 200:
            print('server ready')
            print('API key:', API_KEY)
            break
    except Exception:
        time.sleep(2)

import json
import urllib.request

payload = {
    'model': 'qwen3.8-27b',
    'messages': [
        {'role': 'user', 'content': '用三句话介绍一下你自己。'}
    ],
    'max_tokens': 512,
    'temperature': 0.7,
}

request = urllib.request.Request(
    'http://127.0.0.1:8000/v1/chat/completions',
    data=json.dumps(payload).encode(),
    headers={
        'Content-Type': 'application/json',
        'Authorization': 'Bearer ' + API_KEY,
    },
    method='POST',
)

response = urllib.request.urlopen(request, timeout=90)
result = json.loads(response.read())
print(result['choices'][0]['message']['content'])

import os
import re
import subprocess
import time

if not os.path.exists('/content/cloudflared'):
    subprocess.run([
        'wget', '-q',
        'https://github.com/cloudflare/cloudflared/releases/latest/download/cloudflared-linux-amd64',
        '-O', '/content/cloudflared',
    ], check=True)
    os.chmod('/content/cloudflared', 0o755)

log = open('/content/cloudflared.log', 'w')
subprocess.Popen([
    '/content/cloudflared', 'tunnel', '--no-autoupdate',
    '--url', 'http://127.0.0.1:8000',
], stdout=log, stderr=subprocess.STDOUT)

public_url = None
for _ in range(60):
    text = open('/content/cloudflared.log').read()
    match = re.search(r'https://[a-z0-9-]+\\.trycloudflare\\.com', text)
    if match:
        public_url = match.group(0)
        break
    time.sleep(1)

print('Public API:', public_url + '/v1')

request = urllib.request.Request(
    public_url + '/v1/models',
    headers={'Authorization': 'Bearer ' + API_KEY},
)
response = urllib.request.urlopen(request, timeout=30)
print(response.status)
print(response.read().decode())

from openai import OpenAI

client = OpenAI(
    base_url='https://你的隧道地址.trycloudflare.com/v1',
    api_key='你的 API key',
    timeout=120,
)

stream = client.chat.completions.create(
    model='qwen3.8-27b',
    messages=[{'role': 'user', 'content': '你好。'}],
    max_tokens=2048,
    stream=True,
)

for chunk in stream:
    if not chunk.choices:
        continue
    print(chunk.choices[0].delta.content or '', end='', flush=True)

from openai import OpenAI

client = OpenAI(
    base_url='https://你的隧道地址.trycloudflare.com/v1',
    api_key='从 Drive 凭据文件读取',
    timeout=120,
)

stream = client.chat.completions.create(
    model='qwen3.8-27b',
    messages=[
        {'role': 'user', 'content': '写一段关于远程 GPU 的短文。'}
    ],
    max_tokens=2048,
    stream=True,
)

for chunk in stream:
    if not chunk.choices:
        continue
    print(chunk.choices[0].delta.content or '', end='', flush=True)

import subprocess
import time

socket_path = '/tmp/tailscaled.sock'
subprocess.Popen([
    'sudo', 'tailscaled',
    '--tun=userspace-networking',
    f'--socket={socket_path}',
    '--state=/tmp/tailscale-state',
], stdout=open('/tmp/tailscaled.log', 'w'),
   stderr=subprocess.STDOUT)
time.sleep(3)

from google.colab import userdata

auth_key = userdata.get('TAILSCALE_AUTH_KEY')
subprocess.run([
    'sudo', 'tailscale', f'--socket={socket_path}', 'up',
    f'--authkey={auth_key}',
    '--hostname=colab-qwen',
    '--accept-routes=false',
], check=True)

import time
import urllib.request

while True:
    try:
        print(time.strftime('%H:%M'), urllib.request.urlopen(
            'http://127.0.0.1:8000/health', timeout=10
        ).status, flush=True)
    except Exception as e:
        print('server error:', e, flush=True)
    time.sleep(240)