chore: increase submit limit from 2 to 5

main.py

@@ -31,7 +31,7 @@ PORT = 8080
 STRATEGY_ID = os.getenv("STRATEGY_ID", "")
 
 # 目标GPU
-TARGET_GPU = "Iluvatar_bi-150"
+TARGET_GPU = "Iluvatar_bi-100"
 
 # 账号Token
 TARGET_TOKEN = "f45f1aae2c094426be237c88b1085015"
@@ -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', 'Qwen1.5', 'Qwen-']
+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}"
 
@@ -254,7 +247,7 @@ def check_queue_available() -> int:
 
 
 def build_config_params() -> str:
-    """构建 YAML 配置 - 完全匹配平台自动生成的格式（只支持vllm）"""
+    """构建 YAML 配置 - vllm"""
     params = {
         'framework': 'vllm',
         'nv_framework': 'vllm',
@@ -321,7 +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)
@@ -347,11 +340,11 @@ def run_pipeline(submit_limit: int = 2):
         time.sleep(0.3)
     log(f"搜索完成: {len(seen)} 个唯一模型, {len(all_models)} 个下载量{DOWNLOAD_MIN}-{DOWNLOAD_MAX}")
 
-    # 2. 格式筛选（排除 GGUF/GPTQ/AWQ）
+    # 2. 格式筛选（排除 GPTQ/AWQ，保留 GGUF 和 HuggingFace）
     log("\n--- 阶段2: 格式筛选 ---")
     hf_models = []
     format_skipped = 0
-    SKIP_FORMATS = ['GGUF', 'GPTQ', 'AWQ']
+    SKIP_FORMATS = ['GPTQ', 'AWQ']
     for m in all_models:
         mid_upper = m['model_id'].upper()
         skip = False
@@ -363,27 +356,24 @@ def run_pipeline(submit_limit: int = 2):
                 break
         if not skip:
             hf_models.append(m)
-    log(f"格式筛选: {len(hf_models)} 通过, {format_skipped} 跳过 (GGUF/GPTQ/AWQ)")
+    log(f"格式筛选: {len(hf_models)} 通过, {format_skipped} 跳过 (GPTQ/AWQ)")
 
-    # 3. 架构筛选（排除 Qwen3.5 在 Iluvatar 上不支持）
-    log("\n--- 阶段3: 架构筛选 ---")
+    # 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
-    SKIP_ARCHS = ['qwen3_5', 'Qwen3_5', 'Qwen3.5']
     for m in hf_models:
         ok, reason = check_architecture(m['model_id'])
         if not ok:
             arch_rejected += 1
             log(f"  x {m['model_id']}: {reason}")
-            continue
-        arch_str = str(reason).upper()
-        if any(a.upper() in arch_str for a in SKIP_ARCHS):
+        elif reason == 'GGUF':
             arch_rejected += 1
-            log(f"  x {m['model_id']}: Qwen3.5(Iluvatar不支持)")
-            continue
+            log(f"  x {m['model_id']}: GGUF(无config)")
+        else:
             arch_passed.append(m)
-        time.sleep(0.15)
-    log(f"架构筛选: {len(arch_passed)} 通过, {arch_rejected} 拒绝")
+        time.sleep(0.1)
+    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}] ---")
@@ -428,8 +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}")
@@ -496,7 +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})
 
@@ -713,7 +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()}")