mention no save process for chunk step

Author: wangyukai

src/content/docs/evaluation/custom-model.md

@@ -176,7 +176,7 @@ finally:
 
 ## Tips
 
-- **Use chunk mode for better performance**: Submit action lists via `client.step()` rather than single actions. The server executes them internally and returns only the final observation at the re-inference point.
+- **Use chunk mode for better performance**: Submit action lists via `client.step()` rather than single actions. The server executes them internally and returns only the final observation at the re-inference point. Only work for server with --no_save_process flag.
 - Load model weights in `__init__`, keep inference methods focused on prediction.
 - Use `reset=True` in `obs` to detect the first step and clear recurrent/buffer state.
 - Tune `chunk_size` to balance latency and throughput (e.g., `gmp eval --chunk_size 4`).

src/content/docs/evaluation/run-benchmark.md

@@ -6,13 +6,13 @@ description: Start the Isaac Sim server and run an EBench evaluation.
 ## 1. Start the server
 
 ```bash
-python ray_eval_server.py --host 0.0.0.0 --port 8087
+python ray_eval_server.py --host 0.0.0.0 --port 8087 --no_save_process
 ```
 
 Or with a local Isaac Sim installation:
 
 ```bash
-/isaac-sim/python.sh ray_eval_server.py --host 0.0.0.0 --port 8087
+/isaac-sim/python.sh ray_eval_server.py --host 0.0.0.0 --port 8087 --no_save_process
 ```
 
 ## 2. Submit a task