feat: standalone cancel-all version - startup auto cancels all waiting/running tasks

- Target single GPU (ppu_zw_810e) instead of 4 GPUs
- Search only from HuggingFace API (removed ModelScope)
- Framework fixed to vllm (removed llama.cpp)
- Filter out GGUF format models
- Simplified config and functions

main.py

@@ -1,26 +1,18 @@
 """
-ModelHub 全云端提交智能体
-部署在平台容器中，直接从内网提交验证任务
-
-功能：
-- POST /run    → 触发一次完整流程（搜索→筛选→提交）
-- GET /status  → 查看当前状态
-- GET /health  → 健康检查
+ModelHub Cancel-All 智能体
+启动后自动取消当前账号所有 waiting/running 状态的验证任务
+用法：推送此版本到平台，选择对应 tag 运行即可
 """
 
 import json
 import os
 import signal
-import sqlite3
-import threading
 import time
-import traceback
+import threading
 from datetime import datetime
 from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
-from urllib.parse import urlparse, parse_qs
 
 import requests
-import yaml
 
 # ============================================================
 # 配置
@@ -28,60 +20,19 @@ import yaml
 
 HOST = "0.0.0.0"
 PORT = 8080
-STRATEGY_ID = os.getenv("STRATEGY_ID", "")
 
-# 四个账号的 Token
-ACCOUNTS = {
-    'MetaX_c-500': 'f8e60d1dac7f4472967e7ca40145747b',
-    'Kunlunxin_p-800': 'f45f1aae2c094426be237c88b1085015',
-    'Ascend_910-b4': 'f3c05879e7c34bbba92f399f12884183',
-    'hygon_k100-ai': 'b88507029b884ad3b4bad8ba09e6546e',
-}
+# 目标GPU和Token（跟提交版本保持一致）
+TARGET_GPU = "Iluvatar_bi-150"
+TARGET_TOKEN = "f45f1aae2c094426be237c88b1085015"
 
-# GPU 引擎配置
-GPU_CONFIGS = {
-    'MetaX_c-500': {
-        'framework': 'vllm',
-        'docker_image': 'modelhubxc-4pd.tencentcloudcr.com/enginex/enginex-metax/vllm:0.9.1',
-    },
-    'Kunlunxin_p-800': {
-        'framework': 'vllm',
-        'docker_image': 'modelhubxc-4pd.tencentcloudcr.com/enginex/sunjichen/xc-llm-kunlun:latest',
-    },
-    'Ascend_910-b4': {
-        'framework': 'vllm',
-        'docker_image': 'git.modelhub.org.cn:9443/enginex-ascend/vllm-ascend:v0.11.0rc0',
-    },
-    'hygon_k100-ai': {
-        'framework': 'llama.cpp',
-        'docker_image': 'modelhubxc-4pd.tencentcloudcr.com/enginex/enginex-hygon/hygon-llama.cpp:b7516',
-    },
-}
-
-# 架构白名单
-SUPPORTED_ARCH_KEYWORDS = ['Qwen', 'Qwen2', 'Qwen3']
-SUPPORTED_MODEL_TYPES = [
-    'qwen', 'qwen2', 'qwen2_vl', 'qwen2_5_vl', 'qwen2_audio',
-    'qwen3', 'qwen3_vl', 'qwen3_5', 'qwen3_5_moe',
-]
-SUPPORTED_SPECIAL_ARCHS = ['Eagle3Speculator', 'LlamaForCausalLMEagle3']
-
-MODELSCOPE_API = "https://modelscope.cn/api/v1"
 MODELHUB_API = "https://modelhub.org.cn/api"
-
-# 搜索关键词
-SEARCH_KEYWORDS = ['qwen', 'Qwen2', 'Qwen3', 'Qwen3.5', 'Qwen1.5', 'Qwen-']
+BATCH_SIZE = 50  # 每批取消数量（API上限）
 
 # ============================================================
 # 全局状态
 # ============================================================
 
-state = {
-    'running': False,
-    'last_run': None,
-    'last_result': None,
-    'logs': [],
-}
+state = {'running': False, 'last_run': None, 'logs': []}
 state_lock = threading.Lock()
 
 
@@ -95,378 +46,76 @@ def log(msg: str):
             state['logs'] = state['logs'][-300:]
 
 
-# ============================================================
-# 数据库（内存 SQLite）
-# ============================================================
-
-db_conn = None
-
-
-def init_db():
-    global db_conn
-    db_conn = sqlite3.connect(':memory:', check_same_thread=False)
-    db_conn.execute('''CREATE TABLE IF NOT EXISTS queue (
-        model_id TEXT, gpu TEXT, url TEXT, downloads INTEGER,
-        params TEXT, category TEXT, score REAL,
-        PRIMARY KEY(model_id, gpu)
-    )''')
-    db_conn.execute('''CREATE TABLE IF NOT EXISTS submitted (
-        model_id TEXT, gpu TEXT, task_id TEXT, submitted_at TEXT,
-        PRIMARY KEY(model_id, gpu)
-    )''')
-    db_conn.commit()
-
-
-# ============================================================
-# ModelScope 搜索
-# ============================================================
-
-def search_models(keyword: str, limit: int = 50) -> list:
-    """从 ModelScope 搜索模型"""
-    url = "https://modelscope.cn/openapi/v1/models"
-    params = {
-        'search': keyword,
-        'page_size': min(limit, 50),  # API 上限 50
-        'page_number': 1,
-        'sort': 'downloads',
-    }
-    try:
-        resp = requests.get(url, params=params, timeout=20,
-                            headers={'User-Agent': 'Mozilla/5.0'})
-        data = resp.json()
-        if data.get('success'):
-            models = data.get('data', {}).get('models', [])
-            log(f"  搜索 [{keyword}]: status={resp.status_code} models={len(models)}")
-            return models
-        else:
-            log(f"  搜索 [{keyword}]: success=false, data={str(data)[:200]}")
-    except Exception as e:
-        log(f"  搜索失败 [{keyword}]: {e}")
-    return []
-
-
-def check_architecture(model_id: str) -> tuple:
-    """检查模型架构"""
-    try:
-        cfg_url = f"{MODELSCOPE_API}/models/{model_id}/repo?Revision=master&FilePath=config.json"
-        resp = requests.get(cfg_url, timeout=10)
-        if resp.status_code == 200:
-            cfg = resp.json()
-            archs = cfg.get('architectures', [])
-            mtype = cfg.get('model_type', '')
-            arch_str = str(archs)
-            for special in SUPPORTED_SPECIAL_ARCHS:
-                if special in arch_str:
-                    return True, special
-            for kw in SUPPORTED_ARCH_KEYWORDS:
-                if kw in arch_str:
-                    return True, kw
-            if mtype in SUPPORTED_MODEL_TYPES:
-                return True, mtype
-            return False, f"arch={archs} type={mtype}"
-        mid_upper = model_id.upper()
-        if 'QWEN3' in mid_upper or 'QWEN2' in mid_upper:
-            return True, "GGUF"
-        return False, "no config, not Qwen"
-    except Exception as e:
-        return True, f"check error: {e}"
-
-
-def normalize_model_url(model_url: str) -> str:
-    """标准化 URL 格式"""
-    if '/models/' in model_url:
-        return model_url
-    if 'modelscope.cn/' in model_url:
-        parts = model_url.split('modelscope.cn/')
-        if len(parts) == 2:
-            return f"https://www.modelscope.cn/models/{parts[1]}"
-    return model_url
-
-
-# ============================================================
-# 平台 API
-# ============================================================
-
-def check_platform_verify(model_id: str) -> dict:
-    """查询全平台验证状态"""
-    headers = {'Xc-Token': list(ACCOUNTS.values())[0], 'Accept': 'application/json'}
-    url = f"{MODELHUB_API}/computility/models/search-by-model-id"
-    try:
-        resp = requests.get(url, headers=headers, params={'modelId': model_id}, timeout=10)
-        data = resp.json()
-        if data.get('code') == 0:
-            result = data.get('data', {}).get('verifyResult', {})
-            return result if isinstance(result, dict) else {}
-    except Exception:
-        pass
-    return {}
-
-
-def check_my_submitted(model_id: str, gpu: str, token: str) -> bool:
-    """检查自己是否已提交"""
-    headers = {'Xc-Token': token, 'Accept': 'application/json'}
-    url = f"{MODELHUB_API}/adapt/task/page"
-    try:
-        resp = requests.get(url, headers=headers, params={
-            'current': 1, 'pageSize': 100, 'onlyMine': 'true',
-            'gpuType': gpu, 'modelId': model_id,
-        }, timeout=10)
-        data = resp.json()
-        if data.get('code') == 0:
-            records = data.get('data', {}).get('records', [])
-            for r in records:
-                if r.get('modelId') == model_id and r.get('gpuType') == gpu:
-                    status = r.get('status', '')
-                    if status in ('waiting', 'running', 'success'):
-                        return True
-    except Exception:
-        pass
-    return False
-
-
-def check_queue_available(gpu: str, token: str) -> int:
-    """查询队列可用位置"""
-    headers = {'Xc-Token': token, 'Accept': 'application/json'}
-    url = f"{MODELHUB_API}/adapt/task/page"
-    try:
-        resp = requests.get(url, headers=headers, params={
-            'current': 1, 'pageSize': 1, 'onlyMine': 'true',
-            'gpuType': gpu, 'status': 'waiting',
-        }, timeout=10)
-        data = resp.json()
-        if data.get('code') == 0:
-            waiting = int(data['data'].get('total', 0))
-            return max(0, 100 - waiting)
-    except Exception:
-        pass
-    return -1
-
-
-def build_config_params(gpu: str) -> str:
-    """构建 YAML 配置"""
-    config = GPU_CONFIGS.get(gpu, {})
-    framework = config.get('framework', 'vllm')
-    docker_image = config.get('docker_image', '')
-
-    if framework == 'llama.cpp':
-        params = {
-            'framework': 'llama.cpp',
-            'docker_image': docker_image,
-            'nv_docker_image': docker_image,
-            'api': 'completion',
-            'max_tokens': 1024,
-            'temperature': 0.7,
-            'repetition_penalty': 1.2,
-            'top_p': 0.9,
-            'lang': 'zh',
-            'max_model_len': 4096,
-            'modelFormat': 'GGUF',
-            'sut_config': {
-                'values': {
-                    'gpu_num': 1,
-                    'command': [
-                        '/bin/bash', '-ic',
-                        'llama-server --model /model --alias llm --threads 20 '
-                        '--n-gpu-layers 999 --prio 3 --min_p 0.01 '
-                        '--ctx-size 4096 --host 0.0.0.0 --port 8000 --jinja --flash-attn off'
-                    ]
-                }
-            },
-            'ref_config': {
-                'values': {
-                    'cpu_num': 2, 'gpu_num': 1,
-                    'command': [
-                        'llama-server', '--model', '/model', '--alias', 'llm',
-                        '--threads', '20', '--n-gpu-layers', '999',
-                        '--ctx-size', '4096', '--host', '0.0.0.0', '--port', '8000',
-                    ]
-                }
-            },
-            'model': 'llm',
-        }
-    else:
-        params = {
-            'framework': 'vllm',
-            'docker_image': docker_image,
-            'nv_docker_image': docker_image,
-            'api': 'completion',
-            'max_tokens': 1024,
-            'temperature': 0.7,
-            'repetition_penalty': 1.2,
-            'top_p': 0.9,
-            'lang': 'zh',
-            'max_model_len': 2048,
-            'modelFormat': 'HuggingFace',
-            'sut_config': {
-                'values': {
-                    'gpu_num': 1,
-                    'command': [
-                        '/bin/bash', '-ic',
-                        'vllm serve /model --port 8000 --served-model-name llm '
-                        '--max-model-len 2048 --dtype auto --gpu-memory-utilization 0.95 '
-                        '-tp 1 --enforce-eager --trust-remote-code'
-                    ]
-                }
-            },
-            'ref_config': {
-                'values': {
-                    'cpu_num': 2, 'gpu_num': 1,
-                    'command': [
-                        'vllm', 'serve', '/model', '--port', '8000',
-                        '--served-model-name', 'llm', '--max-model-len', '2048',
-                        '--dtype', 'auto', '--gpu-memory-utilization', '0.95',
-                        '-tp', '1', '--enforce-eager', '--trust-remote-code',
-                    ]
-                }
-            },
-            'model': 'llm',
-        }
-    return yaml.dump(params, default_flow_style=False, allow_unicode=True)
-
-
-def submit_model(model_url: str, gpu: str, token: str) -> tuple:
-    """提交单个模型"""
+def cancel_all_tasks():
+    """取消所有 waiting 和 running 状态的任务"""
     headers = {
-        'Xc-Token': token,
+        'Xc-Token': TARGET_TOKEN,
         'Accept': 'application/json',
         'Content-Type': 'application/json',
     }
-    url = f"{MODELHUB_API}/adapt/task/add"
-    config = GPU_CONFIGS.get(gpu, {})
-    payload = {
-        'modelAddress': normalize_model_url(model_url),
-        'taskType': 'text-generation',
-        'targetGpu': gpu,
-        'framework': config.get('framework', 'vllm'),
-        'strategyId': STRATEGY_ID,
-        'configParams': build_config_params(gpu),
-    }
+
+    log("=" * 50)
+    log("开始取消所有任务")
+    log(f"目标GPU: {TARGET_GPU}")
+
+    # 1. 查询所有任务
+    all_tasks = []
+    page = 1
+    while True:
         try:
-        resp = requests.post(url, headers=headers, json=payload, timeout=30)
+            resp = requests.get(
+                f"{MODELHUB_API}/adapt/task/page",
+                headers={'Xc-Token': TARGET_TOKEN, 'Accept': 'application/json'},
+                params={'current': page, 'pageSize': 100, 'onlyMine': 'true',
+                        'gpuType': TARGET_GPU},
+                timeout=15
+            )
+            data = resp.json()
+            if data.get('code') != 0:
+                log(f"查询失败: code={data.get('code')}")
+                break
+            records = data['data'].get('records', [])
+            if not records:
+                break
+            all_tasks.extend(records)
+            total = data['data'].get('total', 0)
+            log(f"  第{page}页: {len(records)} 条, 累计 {len(all_tasks)} / {total}")
+            if len(all_tasks) >= int(total):
+                break
+            page += 1
+        except Exception as e:
+            log(f"  查询异常: {e}")
+            return
+
+    # 2. 筛选 waiting/running 状态
+    cancellable = [t for t in all_tasks if t.get('status') in ('waiting', 'running')]
+    if not cancellable:
+        log(f"  没有可取消的任务 (总任务 {len(all_tasks)} 个)")
+        log("=" * 50)
+        return
+
+    task_ids = [int(t.get('taskId')) for t in cancellable if t.get('taskId')]
+    log(f"  总任务 {len(all_tasks)} 个, 可取消 {len(task_ids)} 个")
+
+    # 3. 分批取消
+    url = f"{MODELHUB_API}/async/task/stop-create-contest-task"
+    cancelled = 0
+    for i in range(0, len(task_ids), BATCH_SIZE):
+        batch = task_ids[i:i + BATCH_SIZE]
+        try:
+            resp = requests.put(url, headers=headers, json={'taskIds': batch}, timeout=15)
             data = resp.json()
             if data.get('code') == 0:
-            task_id = data.get('data', {}).get('id')
-            return True, task_id, 'success'
-        return False, None, data.get('message', 'unknown error')
+                cancelled += len(batch)
+                log(f"  批次 {i // BATCH_SIZE + 1}: 取消 {len(batch)} 个 OK")
+            else:
+                log(f"  批次 {i // BATCH_SIZE + 1}: 失败 code={data.get('code')} msg={data.get('message', '')[:80]}")
         except Exception as e:
-        return False, None, str(e)
+            log(f"  批次 {i // BATCH_SIZE + 1}: 异常 {e}")
+        time.sleep(1)
 
-
-# ============================================================
-# 主流程
-# ============================================================
-
-def run_pipeline(gpus: list = None, submit_limit: int = 30):
-    """完整流程：搜索→筛选→提交"""
-    if gpus is None:
-        gpus = list(GPU_CONFIGS.keys())
-
-    init_db()
+    log(f"  取消完成: {cancelled} / {len(task_ids)}")
     log("=" * 50)
-    log("开始执行流程")
-    log(f"目标GPU: {', '.join(gpus)}")
-    log(f"提交限制: 每GPU {submit_limit} 个")
-
-    # 1. 搜索
-    log("\n--- 阶段1: 搜索 ModelScope ---")
-    seen = set()
-    all_models = []
-    for kw in SEARCH_KEYWORDS:
-        models = search_models(kw, limit=100)
-        for m in models:
-            mid = m.get('id', '')
-            if mid and mid not in seen:
-                seen.add(mid)
-                downloads = m.get('downloads', 0)
-                if downloads >= 50:
-                    all_models.append({
-                        'model_id': mid,
-                        'url': f"https://modelscope.cn/{mid}",
-                        'downloads': downloads,
-                        'params': m.get('params', ''),
-                        'category': 'quantized' if 'GGUF' in mid.upper() else 'standard',
-                    })
-        time.sleep(0.3)
-    log(f"搜索完成: {len(seen)} 个唯一模型, {len(all_models)} 个下载量>=50")
-
-    # 2. 架构筛选
-    log("\n--- 阶段2: 架构筛选 ---")
-    arch_passed = []
-    arch_rejected = 0
-    for m in all_models:
-        ok, reason = check_architecture(m['model_id'])
-        if ok:
-            arch_passed.append(m)
-        else:
-            arch_rejected += 1
-            log(f"  ✗ {m['model_id']}: {reason}")
-        time.sleep(0.15)
-    log(f"架构筛选: {len(arch_passed)} 通过, {arch_rejected} 拒绝")
-
-    # 3. 按 GPU 筛选并提交
-    log("\n--- 阶段3: 筛选并提交 ---")
-    total_submitted = 0
-    for gpu in gpus:
-        token = ACCOUNTS.get(gpu)
-        if not token:
-            continue
-
-        log(f"\n[{gpu}]")
-
-        # 检查队列
-        available = check_queue_available(gpu, token)
-        if available <= 0:
-            log(f"  队列满，跳过")
-            continue
-        log(f"  队列可用: {available}")
-
-        # 筛选
-        to_submit = []
-        for m in arch_passed:
-            model_id = m['model_id']
-
-            # hygon 只接受 GGUF
-            if gpu == 'hygon_k100-ai' and 'GGUF' not in model_id.upper():
-                continue
-
-            # 检查全平台验证状态
-            verify = check_platform_verify(model_id)
-            if gpu in verify:
-                continue  # 已有记录，跳过
-
-            # 检查自己是否已提交
-            if check_my_submitted(model_id, gpu, token):
-                continue
-
-            to_submit.append(m)
-            if len(to_submit) >= min(submit_limit, available):
-                break
-            time.sleep(0.2)
-
-        log(f"  待提交: {len(to_submit)}")
-
-        # 提交
-        submitted = 0
-        for m in to_submit:
-            ok, task_id, msg = submit_model(m['url'], gpu, token)
-            if ok:
-                submitted += 1
-                log(f"  ✅ {m['model_id']}")
-                db_conn.execute(
-                    'INSERT OR REPLACE INTO submitted VALUES (?,?,?,?)',
-                    (m['model_id'], gpu, str(task_id), datetime.now().isoformat())
-                )
-            else:
-                log(f"  ❌ {m['model_id']}: {msg}")
-            time.sleep(0.5)
-
-        log(f"  提交完成: {submitted}/{len(to_submit)}")
-        total_submitted += submitted
-
-    db_conn.commit()
-    log(f"\n{'=' * 50}")
-    log(f"流程完成，共提交 {total_submitted} 个模型")
-    return total_submitted
 
 
 # ============================================================
@@ -475,151 +124,21 @@ def run_pipeline(gpus: list = None, submit_limit: int = 30):
 
 class AgentHandler(BaseHTTPRequestHandler):
     def do_GET(self):
-        parsed = urlparse(self.path)
-        path = parsed.path
-
-        if path == '/health':
+        if self.path == '/health':
             self._json({'status': 'ok'})
-        elif path == '/':
+        elif self.path == '/':
             self._json({
-                'name': 'modelhub-submit-agent',
-                'strategy_id': STRATEGY_ID,
+                'name': 'modelhub-cancel-all',
+                'gpu': TARGET_GPU,
                 'status': 'running' if state['running'] else 'idle',
-                'last_run': state['last_run'],
-                'last_result': state['last_result'],
+                'last_result': state.get('last_result'),
             })
-        elif path == '/test':
-            self._json(self._run_connectivity_test())
-        elif path == '/status':
-            self._json({
-                'running': state['running'],
-                'last_run': state['last_run'],
-                'last_result': state['last_result'],
-                'queue_count': db_conn.execute('SELECT COUNT(*) FROM queue').fetchone()[0] if db_conn else 0,
-                'submitted_count': db_conn.execute('SELECT COUNT(*) FROM submitted').fetchone()[0] if db_conn else 0,
-            })
-        elif path == '/logs':
-            lines = int(parse_qs(parsed.query).get('lines', ['50'])[0])
+        elif self.path.startswith('/logs'):
+            lines = 100
             self._json({'logs': state['logs'][-lines:]})
         else:
             self._json({'error': 'not found'}, 404)
 
-    def do_POST(self):
-        parsed = urlparse(self.path)
-        path = parsed.path
-
-        if path == '/run':
-            if state['running']:
-                self._json({'error': 'already running'}, 409)
-                return
-
-            # 读取请求体
-            content_len = int(self.headers.get('Content-Length', 0))
-            body = {}
-            if content_len > 0:
-                body = json.loads(self.rfile.read(content_len))
-
-            gpus = body.get('gpus', list(GPU_CONFIGS.keys()))
-            limit = body.get('limit', 30)
-
-            self._json({'status': 'started', 'gpus': gpus, 'limit': limit})
-
-            # 后台运行
-            def _run():
-                try:
-                    state['running'] = True
-                    state['last_run'] = datetime.now().isoformat()
-                    count = run_pipeline(gpus=gpus, submit_limit=limit)
-                    state['last_result'] = {'submitted': count, 'success': True}
-                except Exception as e:
-                    log(f"流程异常: {traceback.format_exc()}")
-                    state['last_result'] = {'error': str(e), 'success': False}
-                finally:
-                    state['running'] = False
-
-            threading.Thread(target=_run, daemon=True).start()
-        else:
-            self._json({'error': 'not found'}, 404)
-
-    def _run_connectivity_test(self) -> dict:
-        """测试各 API 连通性"""
-        results = {}
-
-        # 1. ModelScope 搜索 API
-        try:
-            resp = requests.get(
-                'https://modelscope.cn/openapi/v1/models',
-                params={'search': 'qwen', 'page_size': 2, 'sort': 'downloads'},
-                timeout=10,
-                headers={'User-Agent': 'Mozilla/5.0'}
-            )
-            results['modelscope_search'] = {
-                'status': resp.status_code,
-                'ok': resp.status_code == 200,
-                'body_preview': resp.text[:200] if resp.status_code == 200 else resp.text[:100],
-            }
-        except Exception as e:
-            results['modelscope_search'] = {'ok': False, 'error': str(e)}
-
-        # 2. ModelScope config.json API
-        try:
-            resp = requests.get(
-                'https://modelscope.cn/api/v1/models/Qwen/Qwen3-8B/repo?Revision=master&FilePath=config.json',
-                timeout=10,
-            )
-            results['modelscope_config'] = {
-                'status': resp.status_code,
-                'ok': resp.status_code == 200,
-                'body_preview': resp.text[:200] if resp.status_code == 200 else resp.text[:100],
-            }
-        except Exception as e:
-            results['modelscope_config'] = {'ok': False, 'error': str(e)}
-
-        # 3. ModelHub 查询 API
-        try:
-            token = list(ACCOUNTS.values())[0]
-            resp = requests.get(
-                'https://modelhub.org.cn/api/adapt/task/page',
-                headers={'Xc-Token': token, 'Accept': 'application/json'},
-                params={'current': 1, 'pageSize': 1, 'onlyMine': 'true'},
-                timeout=10,
-            )
-            data = resp.json()
-            results['modelhub_query'] = {
-                'ok': data.get('code') == 0,
-                'code': data.get('code'),
-                'total': data.get('data', {}).get('total'),
-            }
-        except Exception as e:
-            results['modelhub_query'] = {'ok': False, 'error': str(e)}
-
-        # 4. ModelHub 提交 API (dry test)
-        try:
-            token = list(ACCOUNTS.values())[0]
-            resp = requests.post(
-                'https://modelhub.org.cn/api/adapt/task/add',
-                headers={'Xc-Token': token, 'Accept': 'application/json', 'Content-Type': 'application/json'},
-                json={
-                    'modelAddress': 'https://www.modelscope.cn/models/Qwen/Qwen3-8B',
-                    'taskType': 'text-generation',
-                    'targetGpu': 'Kunlunxin_p-800',
-                    'framework': 'vllm',
-                    'strategyId': STRATEGY_ID,
-                    'configParams': 'framework: vllm\n',
-                },
-                timeout=10,
-            )
-            data = resp.json()
-            results['modelhub_submit'] = {
-                'ok': data.get('code') == 0,
-                'code': data.get('code'),
-                'message': data.get('message', '')[:100],
-            }
-        except Exception as e:
-            results['modelhub_submit'] = {'ok': False, 'error': str(e)}
-
-        return results
-
     def _json(self, body: dict, status: int = 200):
         payload = json.dumps(body, ensure_ascii=False).encode()
         self.send_response(status)
@@ -629,7 +148,7 @@ class AgentHandler(BaseHTTPRequestHandler):
         self.wfile.write(payload)
 
     def log_message(self, fmt, *args):
-        pass  # 静默 HTTP 日志
+        pass
 
 
 # ============================================================
@@ -649,107 +168,27 @@ def main():
     signal.signal(signal.SIGTERM, _handle_signal)
     signal.signal(signal.SIGINT, _handle_signal)
 
-    init_db()
-
     server = ThreadingHTTPServer((HOST, PORT), AgentHandler)
     server.timeout = 1
 
-    log(f"智能体启动 | {HOST}:{PORT}")
-    log(f"STRATEGY_ID: {STRATEGY_ID}")
-    log(f"GPU: {', '.join(GPU_CONFIGS.keys())}")
+    log(f"Cancel-All 智能体启动 | {HOST}:{PORT}")
+    log(f"目标GPU: {TARGET_GPU}")
 
-    # 启动后自动运行连通性测试
-    def _startup_test():
+    # 启动后自动取消所有任务
+    def _auto_cancel():
         time.sleep(2)
-        log("=" * 50)
-        log("连通性测试开始")
-        log("=" * 50)
-
-        # 1. ModelScope 搜索
-        try:
-            resp = requests.get(
-                'https://modelscope.cn/openapi/v1/models',
-                params={'search': 'qwen', 'page_size': 2, 'sort': 'downloads'},
-                timeout=10, headers={'User-Agent': 'Mozilla/5.0'}
-            )
-            if resp.status_code == 200:
-                data = resp.json()
-                models = data.get('data', {}).get('models', [])
-                log(f"[ModelScope搜索] ✅ status={resp.status_code} models={len(models)}")
-                for m in models[:2]:
-                    log(f"  {m.get('id')} downloads={m.get('downloads')}")
-            else:
-                log(f"[ModelScope搜索] ❌ status={resp.status_code} body={resp.text[:100]}")
-        except Exception as e:
-            log(f"[ModelScope搜索] ❌ error={e}")
-
-        # 2. ModelScope config.json
-        try:
-            resp = requests.get(
-                'https://modelscope.cn/api/v1/models/Qwen/Qwen3-8B/repo?Revision=master&FilePath=config.json',
-                timeout=10
-            )
-            if resp.status_code == 200:
-                cfg = resp.json()
-                log(f"[ModelScope配置] ✅ arch={cfg.get('architectures')} type={cfg.get('model_type')}")
-            else:
-                log(f"[ModelScope配置] ❌ status={resp.status_code}")
-        except Exception as e:
-            log(f"[ModelScope配置] ❌ error={e}")
-
-        # 3. ModelHub 查询
-        try:
-            token = list(ACCOUNTS.values())[0]
-            resp = requests.get(
-                'https://modelhub.org.cn/api/adapt/task/page',
-                headers={'Xc-Token': token, 'Accept': 'application/json'},
-                params={'current': 1, 'pageSize': 1, 'onlyMine': 'true'},
-                timeout=10
-            )
-            data = resp.json()
-            if data.get('code') == 0:
-                log(f"[ModelHub查询] ✅ total={data['data'].get('total')}")
-            else:
-                log(f"[ModelHub查询] ❌ code={data.get('code')} msg={data.get('message','')[:80]}")
-        except Exception as e:
-            log(f"[ModelHub查询] ❌ error={e}")
-
-        # 4. ModelHub 提交
-        try:
-            token = list(ACCOUNTS.values())[0]
-            resp = requests.post(
-                'https://modelhub.org.cn/api/adapt/task/add',
-                headers={'Xc-Token': token, 'Accept': 'application/json', 'Content-Type': 'application/json'},
-                json={
-                    'modelAddress': 'https://www.modelscope.cn/models/Qwen/Qwen3-8B',
-                    'taskType': 'text-generation', 'targetGpu': 'Kunlunxin_p-800',
-                    'framework': 'vllm', 'strategyId': STRATEGY_ID,
-                    'configParams': 'framework: vllm\n',
-                }, timeout=10
-            )
-            data = resp.json()
-            log(f"[ModelHub提交] code={data.get('code')} msg={data.get('message','')[:80]}")
-        except Exception as e:
-            log(f"[ModelHub提交] ❌ error={e}")
-
-        log("=" * 50)
-        log("连通性测试完成")
-        log("=" * 50)
-
-        # 开始正式提交流程
-        log("\n自动触发提交流程...")
         try:
             state['running'] = True
             state['last_run'] = datetime.now().isoformat()
-            count = run_pipeline(submit_limit=30)
-            state['last_result'] = {'submitted': count, 'success': True}
+            cancel_all_tasks()
+            state['last_result'] = {'success': True}
         except Exception as e:
-            log(f"流程异常: {traceback.format_exc()}")
-            state['last_result'] = {'error': str(e), 'success': False}
+            log(f"异常: {e}")
+            state['last_result'] = {'error': str(e)}
         finally:
             state['running'] = False
 
-    threading.Thread(target=_startup_test, daemon=True).start()
+    threading.Thread(target=_auto_cancel, daemon=True).start()
 
     while not shutdown_requested:
         server.handle_request()