KAPO modernization rollout

.env

@@ -20,8 +20,8 @@ KAPO_LAZY_EMBED_STARTUP=1
 KAPO_LAZY_MODEL_STARTUP=1
 KAPO_PATCH_BUNDLE_URL=https://drive.google.com/uc?export=download&id=16rIe05GZihhAz7ba8E-WibKaJKbh9eu1
 KAPO_PATCH_MANIFEST_URL=https://drive.google.com/uc?export=download&id=1jLuPMCA3hp9qstZZtpBNzTK0XOmLrV8b
-KAPO_REMOTE_ENV_PASSWORD_B64=M2pPZ3FSYWlIcE80SHhTQmI2TTRyU2tVQzE0bVBYNzE
-KAPO_REMOTE_ENV_URL_B64=aHR0cHM6Ly9kcml2ZS5nb29nbGUuY29tL3VjP2V4cG9ydD1kb3dubG9hZCZpZD0xSjhXYzAwR190UEgwZDNubkN3X19VMkUyRUFPLXY0MGI
+KAPO_REMOTE_ENV_PASSWORD_B64=Wml4VURXdlRyU2FXQ011czNxODF5dEdVNmZ1Ujh3VmY
+KAPO_REMOTE_ENV_URL_B64=aHR0cHM6Ly9kcml2ZS5nb29nbGUuY29tL3VjP2V4cG9ydD1kb3dubG9hZCZpZD0xYlhYeWJaQUpqb3lQeGIxNUJDUXhRYmpJb3VQRHh4TU4
 KAPO_SHARED_STATE_BACKEND=google_drive
 MODEL_PROFILE_ID=hf-debugger-qwen25-7b-instruct
 MODEL_REPO=Qwen/Qwen2.5-1.5B-Instruct

bootstrap_space_runtime.py

@@ -28,8 +28,8 @@ DEFAULT_ENV = {
   "KAPO_LAZY_MODEL_STARTUP": "1",
   "KAPO_PATCH_BUNDLE_URL": "https://drive.google.com/uc?export=download&id=16rIe05GZihhAz7ba8E-WibKaJKbh9eu1",
   "KAPO_PATCH_MANIFEST_URL": "https://drive.google.com/uc?export=download&id=1jLuPMCA3hp9qstZZtpBNzTK0XOmLrV8b",
-  "KAPO_REMOTE_ENV_PASSWORD_B64": "M2pPZ3FSYWlIcE80SHhTQmI2TTRyU2tVQzE0bVBYNzE",
-  "KAPO_REMOTE_ENV_URL_B64": "aHR0cHM6Ly9kcml2ZS5nb29nbGUuY29tL3VjP2V4cG9ydD1kb3dubG9hZCZpZD0xSjhXYzAwR190UEgwZDNubkN3X19VMkUyRUFPLXY0MGI",
+  "KAPO_REMOTE_ENV_PASSWORD_B64": "Wml4VURXdlRyU2FXQ011czNxODF5dEdVNmZ1Ujh3VmY",
+  "KAPO_REMOTE_ENV_URL_B64": "aHR0cHM6Ly9kcml2ZS5nb29nbGUuY29tL3VjP2V4cG9ydD1kb3dubG9hZCZpZD0xYlhYeWJaQUpqb3lQeGIxNUJDUXhRYmpJb3VQRHh4TU4",
   "KAPO_SHARED_STATE_BACKEND": "google_drive",
   "MODEL_PROFILE_ID": "hf-debugger-qwen25-7b-instruct",
   "MODEL_REPO": "Qwen/Qwen2.5-1.5B-Instruct",

brain_server/api/main.py

@@ -269,9 +269,17 @@ def _load_embed_model() -> None:
 
 
 def _load_default_model() -> None:
-    repo_id = os.getenv("MODEL_REPO", DEFAULT_MODEL_REPO)
-    filename = os.getenv("MODEL_FILE", DEFAULT_MODEL_FILE)
-    ensure_model_loaded(repo_id, filename, hf_token=os.getenv("HF_TOKEN"))
+    global MODEL, MODEL_ERROR, MODEL_META
+    repo_id = str(os.getenv("MODEL_REPO", DEFAULT_MODEL_REPO) or DEFAULT_MODEL_REPO).strip()
+    filename = str(os.getenv("MODEL_FILE", "") or "").strip()
+    provider = str(os.getenv("BRAIN_PROVIDER", "") or os.getenv("BRAIN_TEMPLATE", "") or "").strip().lower()
+    if not filename and (_feature_enabled("KAPO_HF_INFERENCE_API", default=False) or "huggingface" in provider or "hf-space" in provider):
+        MODEL = None
+        MODEL_ERROR = None
+        MODEL_META = {"repo_id": repo_id, "filename": "", "path": None}
+        logger.info("Skipping local model load; using Hugging Face inference API for %s", repo_id)
+        return
+    ensure_model_loaded(repo_id, filename or DEFAULT_MODEL_FILE, hf_token=os.getenv("HF_TOKEN"))
 
 
 def _brain_headers() -> dict:
@@ -1609,19 +1617,18 @@ def _generate_response(user_input: str, history: list[dict[str, str]], context_b
                 from huggingface_hub import InferenceClient
                 prompt = _build_chat_prompt(user_input, history, context_block)
                 max_tokens = 80 if language == "ar" else 96
-                client = InferenceClient(api_key=(str(os.getenv("HF_TOKEN", "") or "").strip() or None))
+                model_repo = str(os.getenv("MODEL_REPO", DEFAULT_MODEL_REPO) or DEFAULT_MODEL_REPO).strip()
+                client = InferenceClient(model=model_repo, api_key=(str(os.getenv("HF_TOKEN", "") or "").strip() or None))
                 generated = client.text_generation(
                     prompt,
-                    model=str(os.getenv("MODEL_REPO", DEFAULT_MODEL_REPO) or DEFAULT_MODEL_REPO).strip(),
                     max_new_tokens=max_tokens,
-                    temperature=0.1,
                     return_full_text=False,
                 )
                 generated_text = str(generated or "").strip()
                 if generated_text:
                     return generated_text
-            except Exception:
-                logger.warning("HF inference fallback failed", exc_info=True)
+            except Exception as exc:
+                logger.warning("HF inference fallback failed: %s", exc, exc_info=True)
         if language == "ar":
             return "الخدمة تعمل لكن توليد الرد الحر غير متاح الآن لأن النموذج غير محمل."
         return "The Brain is online, but natural chat generation is unavailable because the model is not loaded."

kapo.env

@@ -20,8 +20,8 @@ KAPO_LAZY_EMBED_STARTUP=1
 KAPO_LAZY_MODEL_STARTUP=1
 KAPO_PATCH_BUNDLE_URL=https://drive.google.com/uc?export=download&id=16rIe05GZihhAz7ba8E-WibKaJKbh9eu1
 KAPO_PATCH_MANIFEST_URL=https://drive.google.com/uc?export=download&id=1jLuPMCA3hp9qstZZtpBNzTK0XOmLrV8b
-KAPO_REMOTE_ENV_PASSWORD_B64=M2pPZ3FSYWlIcE80SHhTQmI2TTRyU2tVQzE0bVBYNzE
-KAPO_REMOTE_ENV_URL_B64=aHR0cHM6Ly9kcml2ZS5nb29nbGUuY29tL3VjP2V4cG9ydD1kb3dubG9hZCZpZD0xSjhXYzAwR190UEgwZDNubkN3X19VMkUyRUFPLXY0MGI
+KAPO_REMOTE_ENV_PASSWORD_B64=Wml4VURXdlRyU2FXQ011czNxODF5dEdVNmZ1Ujh3VmY
+KAPO_REMOTE_ENV_URL_B64=aHR0cHM6Ly9kcml2ZS5nb29nbGUuY29tL3VjP2V4cG9ydD1kb3dubG9hZCZpZD0xYlhYeWJaQUpqb3lQeGIxNUJDUXhRYmpJb3VQRHh4TU4
 KAPO_SHARED_STATE_BACKEND=google_drive
 MODEL_PROFILE_ID=hf-debugger-qwen25-7b-instruct
 MODEL_REPO=Qwen/Qwen2.5-1.5B-Instruct