diff --git a/README.md b/README.md
index a086c93fada95d65a07d1c31d4a8c15dcbd4fa18..75b3f09d2b82f441724ba8c7e4fd67089c2078cf 100644
--- a/README.md
+++ b/README.md
@@ -128,6 +128,8 @@ size的动态变化，相比于静态shape的整图下沉，动态shape的计算
 | OpenAI o1-mini         | 68.9      |
 | DeepSeek R1            | 91.8      |
 | Deepseek R1 w8a8       | 89.52     |
+| Deepseek R1 W4A16      | 88.78     |
+| Deepseek V3 0324 W4A16 | 87.82     |
 
 
 
@@ -141,8 +143,8 @@ size的动态变化，相比于静态shape的整图下沉，动态shape的计算
 
 | 并发数 | 吞吐(Token/s) |
 | ------ | ------------- |
-| 1      | 16.7          |
-| 192    | 1188          |
+| 1      | 22.4          |
+| 192    | 1600          |
 
 
 
diff --git "a/doc/deepseek/DeepSeek-V3&R1\351\203\250\347\275\262\346\214\207\345\215\227.md" "b/doc/deepseek/DeepSeek-V3&R1\351\203\250\347\275\262\346\214\207\345\215\227.md"
index dd60db33b5b28dd7fbfa98d3bdca08a864ab95e2..5515b29c83bbac5c040e72dd4c12708989aa1272 100644
--- "a/doc/deepseek/DeepSeek-V3&R1\351\203\250\347\275\262\346\214\207\345\215\227.md"
+++ "b/doc/deepseek/DeepSeek-V3&R1\351\203\250\347\275\262\346\214\207\345\215\227.md"
@@ -120,6 +120,7 @@ cd llm_solution/script/mindspore-deepseek
 
 ```shell
 # 调整mindspore-deepseek目录下config.yaml
+# 注意：w8a8与int4不同权重部署方式，使用的镜像tag不一致，可按照下文修改
 (base) [root@910b-3 mindspore-deepseek]# cat config.yaml 
 all:
   children:
@@ -142,8 +143,8 @@ all:
   vars:
     # 容器镜像
     # 如果本地docker上已加载镜像，改成docker image的image_name和image_tag
-    image_name: hub.oepkgs.net/oedeploy/openeuler/aarch64/deepseek_hyperinfer
-    image_tag: openeuler22.03-py3.11
+    image_name: hub.oepkgs.net/oedeploy/openeuler/aarch64/mindspore
+    image_tag: 20250415  # 单机部署时，请使用tag为20250326的容器镜像
     # 将要拉起的推理容器的名称
     container_name: openeuler_ds  # 启动之后的docker name，不能和已有镜像重名
     # 模型路径
diff --git a/script/mindspore-deepseek/workspace/roles/prepare/files/lib/start_ds.sh b/script/mindspore-deepseek/workspace/roles/prepare/files/lib/start_ds.sh
index 49a7bd29ecd31b330e1860887a706bc75bba674f..89209c4639029fe411f752f24a01cbf64b314a19 100644
--- a/script/mindspore-deepseek/workspace/roles/prepare/files/lib/start_ds.sh
+++ b/script/mindspore-deepseek/workspace/roles/prepare/files/lib/start_ds.sh
@@ -39,7 +39,7 @@ rm -rf ds.log
 if [ $NODE_NUM -ne 1 ]; then
     nohup python3 -m vllm_mindspore.entrypoints vllm.entrypoints.openai.api_server --model "$MODEL_PATH" --port=$LLM_PORT --trust_remote_code --tensor_parallel_size=$PARALLEL --max-num-seqs=192 --max_model_len=32768 --max-num-batched-tokens=16384 --block-size=128 --gpu-memory-utilization=0.93 --num-scheduler-steps=8 --disable-async-output-proc --distributed-executor-backend=ray &> ds.log &
 else
-    nohup python3 -m vllm_mindspore.entrypoints vllm.entrypoints.openai.api_server --model "$MODEL_PATH" --port=$LLM_PORT --trust_remote_code --tensor_parallel_size=8 --max-num-seqs=192 --max_model_len=32768 --max-num-batched-tokens=16384 --block-size=128 --gpu-memory-utilization=0.93 --num-scheduler-steps=8 --disable-async-output-proc &> ds.log &
+    nohup python3 -m vllm_mindspore.entrypoints vllm.entrypoints.openai.api_server --model "$MODEL_PATH" --port=$LLM_PORT --trust_remote_code --tensor_parallel_size=8 --max-num-seqs=192 --max_model_len=32768 --max-num-batched-tokens=16384 --block-size=32 --gpu-memory-utilization=0.93 --num-scheduler-steps=8 --disable-async-output-proc &> ds.log &
 fi
 
 #检测推理服务是否拉起