Cache size to 0 to improve performance on LLMs (#490)

Author: ezelanza

ai_ref_kits/agentic_multimodal_travel_planer/download_and_run_models_Windows.bat

@@ -93,7 +93,7 @@ REM Start LLM service
 REM --port = gRPC, --rest_port = HTTP REST (chat/completions). Agents use HTTP, so REST must be on LLM_PORT.
 echo Starting LLM service (REST on %LLM_PORT%, gRPC on 8011)...
 set LLM_GRPC_PORT=8011
-set LLM_ARGS=--port %LLM_GRPC_PORT% --rest_port %LLM_PORT% --model_repository_path "%MODELS_DIR%" --source_model "%LLM_MODEL%" --tool_parser hermes3 --cache_size 2 --task text_generation --enable_prefix_caching true
+set LLM_ARGS=--port %LLM_GRPC_PORT% --rest_port %LLM_PORT% --model_repository_path "%MODELS_DIR%" --source_model "%LLM_MODEL%" --tool_parser hermes3 --cache_size 0 --task text_generation 
 if not "%TARGET_DEVICE%"=="" set LLM_ARGS=%LLM_ARGS% --target_device %TARGET_DEVICE%
 REM Use PowerShell Start-Process to launch detached
 powershell -Command "Start-Process -FilePath '%OVMS_PATH%' -ArgumentList '%LLM_ARGS%' -RedirectStandardOutput '%LOGS_DIR%\ovms_llm.log' -RedirectStandardError '%LOGS_DIR%\ovms_llm.err' -WindowStyle Hidden" || (echo Failed to start LLM service && exit /b 1)

ai_ref_kits/agentic_multimodal_travel_planer/download_and_run_models_linux.sh

@@ -427,7 +427,6 @@ if ! docker run -d \
   --source_model "${LLM_MODEL}" \
   --task text_generation \
   --tool_parser ${LLM_TOOL_PARSER} \
-  ${LLM_REASONING_PARSER:+--reasoning_parser ${LLM_REASONING_PARSER}} \
   --log_level DEBUG \
   ${LLM_TARGET_DEVICE_ARG} \
   >/dev/null; then