chore: increase submit limit from 2 to 5

main.py

@@ -47,7 +47,7 @@ SUPPORTED_SPECIAL_ARCHS = ['Eagle3Speculator', 'LlamaForCausalLMEagle3']
 MODELHUB_API = "https://modelhub.org.cn/api"
 
 # 搜索关键词
-SEARCH_KEYWORDS = ['qwen', 'Qwen2', 'Qwen3', 'Qwen3.5', 'Llama-3', 'Llama-3.1', 'Mistral', 'DeepSeek']
+SEARCH_KEYWORDS = ['Llama-3', 'Llama-3.1', 'Llama-3.2', 'Meta-Llama', 'Llama-4']
 
 # ============================================================
 # 全局状态
@@ -157,7 +157,7 @@ def search_models(keyword: str) -> list:
 
 
 def check_architecture(model_id: str) -> tuple:
-    """检查模型架构"""
+    """检查模型是否有有效的 config.json（不限架构类型）"""
     try:
         cfg_url = f"{MODELSCOPE_API}/models/{model_id}/repo?Revision=master&FilePath=config.json"
         resp = requests.get(cfg_url, timeout=10)
@@ -165,20 +165,13 @@ def check_architecture(model_id: str) -> tuple:
             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}"
+            if archs or mtype:
+                return True, f"arch={archs} type={mtype}"
+            return False, "empty config"
         mid_upper = model_id.upper()
-        if 'QWEN3' in mid_upper or 'QWEN2' in mid_upper:
-            return True, "GGUF"
-        return False, "no config, not Qwen"
+        if 'LLAMA' in mid_upper:
+            return True, "Llama(GGUF)"
+        return False, "no config"
     except Exception as e:
         return True, f"check error: {e}"
 
@@ -254,26 +247,25 @@ def check_queue_available() -> int:
 
 
 def build_config_params() -> str:
-    """构建 YAML 配置 - llama.cpp (支持 GGUF)"""
+    """构建 YAML 配置 - vllm"""
     params = {
-        'framework': 'llama.cpp',
-        'nv_framework': 'llama.cpp',
+        'framework': 'vllm',
+        'nv_framework': 'vllm',
         'api': 'completion',
         'max_tokens': 1024,
         'temperature': 0.7,
         'repetition_penalty': 1.2,
         'top_p': 0.9,
         'lang': 'zh',
-        'max_model_len': 4096,
+        'max_model_len': 2048,
         'sut_config': {
             'gpu_num': 1,
             'values': {
                 'command': [
-                    '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',
+                    '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',
                 ]
             }
         },
@@ -281,9 +273,9 @@ def build_config_params() -> str:
             'gpu_num': 1,
             'values': {
                 'command': [
-                    'llama-server', '--model', '/model', '--alias', 'llm',
-                    '--threads', '20', '--n-gpu-layers', '999',
-                    '--ctx-size', '4096', '--host', '0.0.0.0', '--port', '8000',
+                    'vllm', 'serve', '/model', '--port', '80',
+                    '--served-model-name', 'llm', '--max-model-len', '4096',
+                    '--enforce-eager', '--trust-remote-code', '-tp', '1',
                 ]
             }
         },
@@ -303,7 +295,7 @@ def submit_model(model_url: str) -> tuple:
         'modelAddress': normalize_model_url(model_url),
         'taskType': 'text-generation',
         'targetGpu': TARGET_GPU,
-        'framework': 'llama.cpp',
+        'framework': 'vllm',
         'strategyId': STRATEGY_ID,
         'configParams': build_config_params(),
     }
@@ -322,7 +314,7 @@ def submit_model(model_url: str) -> tuple:
 # 主流程
 # ============================================================
 
-def run_pipeline(submit_limit: int = 2):
+def run_pipeline(submit_limit: int = 5):
     """完整流程：搜索→筛选→提交（只针对目标GPU）"""
     init_db()
     log("=" * 50)
@@ -366,14 +358,22 @@ def run_pipeline(submit_limit: int = 2):
             hf_models.append(m)
     log(f"格式筛选: {len(hf_models)} 通过, {format_skipped} 跳过 (GPTQ/AWQ)")
 
-    # 3. 架构筛选（参考检查，不做严格过滤，让平台决定兼容性）
+    # 3. 架构筛选（只保留有标准config.json的模型，排除无效格式）
     log("\n--- 阶段3: 架构检查 ---")
+    arch_passed = []
+    arch_rejected = 0
     for m in hf_models:
         ok, reason = check_architecture(m['model_id'])
         if not ok:
-            log(f"  ! {m['model_id']}: {reason} (仍保留)")
+            arch_rejected += 1
+            log(f"  x {m['model_id']}: {reason}")
+        elif reason == 'GGUF':
+            arch_rejected += 1
+            log(f"  x {m['model_id']}: GGUF(无config)")
+        else:
+            arch_passed.append(m)
         time.sleep(0.1)
-    log(f"架构检查: {len(hf_models)} 个模型进入下一阶段")
+    log(f"架构检查: {len(arch_passed)} 通过, {arch_rejected} 拒绝")
 
     # 4. 筛选并提交（只针对目标GPU）
     log(f"\n--- 阶段4: 筛选并提交 [{TARGET_GPU}] ---")
@@ -387,7 +387,7 @@ def run_pipeline(submit_limit: int = 2):
 
     # 筛选
     to_submit = []
-    for m in hf_models:
+    for m in arch_passed:
         model_id = m['model_id']
 
         # 检查全平台验证状态
@@ -418,8 +418,9 @@ def run_pipeline(submit_limit: int = 2):
             submitted += 1
             log(f"  ✅ {m['model_id']}")
             db_conn.execute(
-                'INSERT OR REPLACE INTO submitted VALUES (?,?,?,?)',
-                (m['model_id'], TARGET_GPU, str(task_id), datetime.now().isoformat())
+                'INSERT OR REPLACE INTO submitted VALUES (?,?,?,?,?,?)',
+                (m['model_id'], TARGET_GPU, str(task_id), 'submitted',
+                 datetime.now().isoformat(), None)
             )
         else:
             log(f"  ❌ {m['model_id']}: {msg}")
@@ -486,7 +487,7 @@ class AgentHandler(BaseHTTPRequestHandler):
             if content_len > 0:
                 body = json.loads(self.rfile.read(content_len))
 
-            limit = body.get('limit', 2)
+            limit = body.get('limit', 5)
 
             self._json({'status': 'started', 'gpu': TARGET_GPU, 'limit': limit})
 
@@ -567,7 +568,7 @@ class AgentHandler(BaseHTTPRequestHandler):
                     'modelAddress': 'https://www.modelscope.cn/models/Qwen/Qwen3-8B',
                     'taskType': 'text-generation',
                     'targetGpu': TARGET_GPU,
-                    'framework': 'llama.cpp',
+                    'framework': 'vllm',
                     'strategyId': STRATEGY_ID,
                     'configParams': build_config_params(),
                 },
@@ -685,7 +686,7 @@ def main():
                 json={
                     'modelAddress': 'https://www.modelscope.cn/models/Qwen/Qwen3-8B',
                     'taskType': 'text-generation', 'targetGpu': TARGET_GPU,
-                    'framework': 'llama.cpp', 'strategyId': STRATEGY_ID,
+                    'framework': 'vllm', 'strategyId': STRATEGY_ID,
                     'configParams': build_config_params(),
                 }, timeout=10
             )
@@ -703,7 +704,7 @@ def main():
         try:
             state['running'] = True
             state['last_run'] = datetime.now().isoformat()
-            count = run_pipeline(submit_limit=2)
+            count = run_pipeline(submit_limit=5)
             state['last_result'] = {'submitted': count, 'success': True}
         except Exception as e:
             log(f"流程异常: {traceback.format_exc()}")