feat: multi-page search, filter downloads 50-5000 for cold models

main.py

@@ -31,7 +31,7 @@ PORT = 8080
 STRATEGY_ID = os.getenv("STRATEGY_ID", "")
 
 # 目标GPU
-TARGET_GPU = "ppu_zw_810e"
+TARGET_GPU = "Iluvatar_bi-150"
 
 # 账号Token
 TARGET_TOKEN = "f45f1aae2c094426be237c88b1085015"
@@ -99,15 +99,20 @@ def init_db():
 # ============================================================
 
 MODELSCOPE_API = "https://modelscope.cn/api/v1"
+DOWNLOAD_MIN = 50
+DOWNLOAD_MAX = 5000
+SEARCH_PAGES = 5  # 每个关键词搜5页（50*5=250个结果）
 
 
-def search_models(keyword: str, limit: int = 50) -> list:
-    """从 ModelScope 搜索模型"""
+def search_models(keyword: str) -> list:
+    """从 ModelScope 搜索模型（多页，筛选下载量50-5000的冷门模型）"""
     url = "https://modelscope.cn/openapi/v1/models"
+    models = []
+    for page in range(1, SEARCH_PAGES + 1):
         params = {
             'search': keyword,
-        'page_size': min(limit, 50),
-        'page_number': 1,
+            'page_size': 50,
+            'page_number': page,
             'sort': 'downloads',
         }
         try:
@@ -115,14 +120,21 @@ def search_models(keyword: str, limit: int = 50) -> list:
                                 headers={'User-Agent': 'Mozilla/5.0'})
             data = resp.json()
             if data.get('success'):
-            models = data.get('data', {}).get('models', [])
-            log(f"  [{keyword}]: {len(models)} 个结果")
-            return [{'id': m.get('id'), 'downloads': m.get('downloads', 0)} for m in models]
+                page_models = data.get('data', {}).get('models', [])
+                for m in page_models:
+                    dl = m.get('downloads', 0)
+                    if DOWNLOAD_MIN <= dl <= DOWNLOAD_MAX:
+                        models.append({'id': m.get('id'), 'downloads': dl})
+                if len(page_models) < 50:
+                    break  # 最后一页，不继续
             else:
-            log(f"  [{keyword}]: success=false")
+                break
         except Exception as e:
-        log(f"  [{keyword}]: 失败 {e}")
-    return []
+            log(f"  [{keyword}] page={page}: {e}")
+            break
+        time.sleep(0.3)
+    log(f"  [{keyword}]: {len(models)} 个 (50<={DOWNLOAD_MAX})")
+    return models
 
 
 def check_architecture(model_id: str) -> tuple:
@@ -303,20 +315,18 @@ def run_pipeline(submit_limit: int = 30):
     seen = set()
     all_models = []
     for kw in SEARCH_KEYWORDS:
-        models = search_models(kw, limit=100)
+        models = search_models(kw)
         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,
+                    'downloads': m.get('downloads', 0),
                 })
         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")
+    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（只保留HuggingFace格式，排除GGUF）
     log("\n--- 阶段2: 格式筛选 ---")
@@ -385,7 +395,7 @@ def run_pipeline(submit_limit: int = 30):
             log(f"  ✅ {m['model_id']}")
             db_conn.execute(
                 'INSERT OR REPLACE INTO submitted VALUES (?,?,?,?)',
-                (m['model_id'], gpu, str(task_id), datetime.now().isoformat())
+                (m['model_id'], TARGET_GPU, str(task_id), datetime.now().isoformat())
             )
         else:
             log(f"  ❌ {m['model_id']}: {msg}")