fix: skip LLM and reranker preload to reduce CPU usage

entrypoint.sh

@@ -54,33 +54,11 @@ try:
 except Exception as e:
     print(f'[ENTRYPOINT] ⚠️ Embedding model preload failed: {e}', flush=True)
 
-# 2. Preload LLM Model (llama.cpp)
-llm_provider = os.environ.get('DEFAULT_LLM_PROVIDER') or os.environ.get('LLM_PROVIDER', '')
-if llm_provider.lower() == 'llama_cpp':
-    try:
-        print('[ENTRYPOINT] 📦 Preloading LLM model (llama.cpp)...', flush=True)
-        from hue_portal.chatbot.llm_integration import get_llm_generator
-        llm_gen = get_llm_generator()
-        if llm_gen and hasattr(llm_gen, 'llama_cpp') and llm_gen.llama_cpp:
-            print('[ENTRYPOINT] ✅ LLM model preloaded successfully', flush=True)
-        else:
-            print('[ENTRYPOINT] ⚠️ LLM model not loaded (may load on first request)', flush=True)
-    except Exception as e:
-        print(f'[ENTRYPOINT] ⚠️ LLM model preload failed: {e} (will load on first request)', flush=True)
-else:
-    print(f'[ENTRYPOINT] ⏭️ Skipping LLM preload (provider is {llm_provider or \"not set\"}, not llama_cpp)', flush=True)
-
-# 3. Preload Reranker Model (lazy, but trigger import)
-try:
-    print('[ENTRYPOINT] 📦 Preloading reranker model...', flush=True)
-    from hue_portal.core.reranker import get_reranker
-    reranker = get_reranker()
-    if reranker:
-        print('[ENTRYPOINT] ✅ Reranker model preloaded successfully', flush=True)
-    else:
-        print('[ENTRYPOINT] ⚠️ Reranker model not loaded (may load on first request)', flush=True)
-except Exception as e:
-    print(f'[ENTRYPOINT] ⚠️ Reranker preload failed: {e} (will load on first request)', flush=True)
+# 2. SKIP LLM Preload (CPU optimization - too heavy, will load on first request)
+print('[ENTRYPOINT] ⏭️ Skipping LLM preload (CPU optimization - will load lazily on first request)', flush=True)
+
+# 3. SKIP Reranker Preload (CPU optimization - too heavy, will load on first request)
+print('[ENTRYPOINT] ⏭️ Skipping reranker preload (CPU optimization - will load lazily on first request)', flush=True)
 
 print('[ENTRYPOINT] ✅ Model preload completed', flush=True)  # v2.0-preload-all
 " || log "⚠️ Model preload had errors (models will load on first request)"