From 02ec313eab6adfc02602b75eaf544a1cee78226a Mon Sep 17 00:00:00 2001
From: Guancheng Fu <110874468+gc-fu@users.noreply.github.com>
Date: Mon, 24 Feb 2025 09:59:17 +0800
Subject: [PATCH] Update README.md (#12877)

---
 python/llm/example/GPU/vLLM-Serving/README.md | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/python/llm/example/GPU/vLLM-Serving/README.md b/python/llm/example/GPU/vLLM-Serving/README.md
index 1ef70a39..7c857686 100644
--- a/python/llm/example/GPU/vLLM-Serving/README.md
+++ b/python/llm/example/GPU/vLLM-Serving/README.md
@@ -241,3 +241,10 @@ llm = LLM(model="DeepSeek-R1-Distill-Qwen-7B", # Unquantized model path on disk
 When finish executing, the low-bit model has been saved at `/llm/fp8-model-path`.
 
 Later we can use the option `--low-bit-model-path /llm/fp8-model-path` to use the low-bit model.
+
+
+### 5. Known issues
+
+#### Runtime memory
+
+If runtime memory is a concern, you can set --swap-space 0.5 to reduce memory consumption during execution. The default value for --swap-space is 4, which means that by default, the system reserves 4GB of memory for use when GPU memory is insufficient.