Upload backend/services/v4_complement_client.py with huggingface_hub

backend/services/v4_complement_client.py

@@ -90,36 +90,44 @@ async def get_models_status() -> Dict[str, Any]:
 
 
 async def get_models_list_summary() -> Dict[str, Any]:
-    """Model catalog from v4 — uses /api/models/list?limit=N for total_models count."""
+    """Model catalog from v4 — status for loaded/hf_mode, list for catalog items."""
+    status = await get_models_status()
+    status_payload = (status.get("data") or {}) if status.get("success") else {}
+
     list_result = await _request(
         "GET",
         "/api/models/list",
-        params={"limit": 8},
+        params={"limit": 12},
         cache_key="v4:models_list_summary",
     )
     if list_result.get("success"):
         payload = list_result.get("data") or {}
         models = payload.get("models") or []
+        model_info = payload.get("model_info") or {}
+        loaded_from_list = sum(1 for m in models if m.get("loaded"))
+        loaded = status_payload.get("models_loaded", loaded_from_list)
         return {
             "success": True,
             "total": payload.get("total_models") or payload.get("total") or len(models),
-            "loaded": sum(1 for m in models if m.get("loaded")),
-            "hf_mode": payload.get("hf_mode") or (payload.get("model_info") or {}).get("hf_mode"),
+            "loaded": loaded,
+            "hf_mode": status_payload.get("hf_mode") or model_info.get("hf_mode"),
+            "inference_api_mode": status_payload.get("inference_api_mode"),
             "items": models[:12],
+            "loaded_models": status_payload.get("loaded_models") or [],
             "source": "v4_complement",
         }
 
-    status = await get_models_status()
     if not status.get("success"):
         return status
-    payload = status.get("data") or {}
-    items = payload.get("items") or payload.get("models", {}).get("items", [])
+    items = status_payload.get("items") or status_payload.get("models", {}).get("items", [])
     return {
         "success": True,
-        "total": payload.get("models_total") or len(items),
-        "loaded": payload.get("models_loaded", 0),
-        "hf_mode": payload.get("hf_mode"),
+        "total": status_payload.get("models_total") or len(items),
+        "loaded": status_payload.get("models_loaded", 0),
+        "hf_mode": status_payload.get("hf_mode"),
+        "inference_api_mode": status_payload.get("inference_api_mode"),
         "items": items[:12],
+        "loaded_models": status_payload.get("loaded_models") or [],
         "source": "v4_complement",
     }