diff --git a/configs/inference/amd-vllm.json b/configs/inference/amd-vllm.json
index dcde3557f..318338024 100644
--- a/configs/inference/amd-vllm.json
+++ b/configs/inference/amd-vllm.json
@@ -6,9 +6,9 @@
   "engine_images": [
     {
       "compute_type": "gpu",
-      "image": "opencsghq/amd-vllm:rocm7.0.0_vllm_0.19.0",
-      "driver_version": "7.0",
-      "engine_version": "0.19.0"
+      "image": "opencsghq/amd-vllm:rocm7.2.1_vllm_0.20.1",
+      "driver_version": "7.2.1",
+      "engine_version": "0.20.1"
     }
   ],
   "engine_args": [
@@ -370,13 +370,18 @@
     "Phi4ForCausalLMV",
     "SarvamMLAForCausalLM",
     "SarvamMoEForCausalLM",
-    "TeleChat3ForCausalLM"
+    "TeleChat3ForCausalLM",
+    "DeepseekV4ForCausalLM",
+    "Rnj1ForCausalLM",
+    "HYV3ForCausalLM",
+    "Ministral3ForCausalLM"
   ],
   "tool_call_parsers": {
     "DeepseekR1ForCausalLM": "deepseek_r1",
     "DeepseekV3ForCausalLM": "deepseek_v3",
     "DeepseekV31ForCausalLM": "deepseek_v31",
     "DeepseekV32ForCausalLM": "deepseek_v32",
+    "DeepseekV4ForCausalLM": "deepseek_v4",
     "Ernie4_5ForCausalLM": "ernie45",
     "Glm4ForCausalLM": "glm47",
     "Glm4MoeForCausalLM": "glm45",
diff --git a/configs/inference/vllm.json b/configs/inference/vllm.json
index c0aacba25..c527d0fcd 100644
--- a/configs/inference/vllm.json
+++ b/configs/inference/vllm.json
@@ -6,9 +6,9 @@
   "engine_images": [
     {
       "compute_type": "gpu",
-      "image": "opencsghq/vllm:v0.19.0",
+      "image": "opencsghq/vllm:v0.20.1",
       "driver_version": "13.0",
-      "engine_version": "v0.19.0",
+      "engine_version": "v0.20.1",
       "extra_archs": [
         "AXK1ForCausalLM",
         "ApertusForCausalLM",
@@ -180,7 +180,11 @@
         "UltravoxModel",
         "VoxtralForConditionalGeneration",
         "VoxtralRealtimeGeneration",
-        "WhisperForConditionalGeneration"
+        "WhisperForConditionalGeneration",
+        "DeepseekV4ForCausalLM",
+        "Rnj1ForCausalLM",
+        "HYV3ForCausalLM",
+        "Ministral3ForCausalLM"
       ]
     },
     {
@@ -393,13 +397,18 @@
     "TeleChat2ForCausalLM",
     "TeleFLMForCausalLM",
     "XverseForCausalLM",
-    "Zamba2ForCausalLM"
+    "Zamba2ForCausalLM",
+    "DeepseekV4ForCausalLM",
+    "Rnj1ForCausalLM",
+    "HYV3ForCausalLM",
+    "Ministral3ForCausalLM"
   ],
   "tool_call_parsers": {
     "DeepseekR1ForCausalLM": "deepseek_r1",
     "DeepseekV3ForCausalLM": "deepseek_v3",
     "DeepseekV31ForCausalLM": "deepseek_v31",
     "DeepseekV32ForCausalLM": "deepseek_v32",
+    "DeepseekV4ForCausalLM": "deepseek_v4",
     "Ernie4_5ForCausalLM": "ernie45",
     "Glm4ForCausalLM": "glm47",
     "Glm4MoeForCausalLM": "glm45",
diff --git a/docker/inference/Dockerfile.vllm b/docker/inference/Dockerfile.vllm
index 3454d9aa2..a26695d3a 100644
--- a/docker/inference/Dockerfile.vllm
+++ b/docker/inference/Dockerfile.vllm
@@ -1,4 +1,4 @@
-FROM vllm/vllm-openai:v0.19.0-cu130
+FROM vllm/vllm-openai:v0.20.1
 RUN pip config set global.index-url https://mirrors.aliyun.com/pypi/simple
 RUN pip install --no-cache-dir csghub-sdk==0.7.3 ray supervisor
 RUN sed -i 's|http://archive.ubuntu.com/ubuntu|http://mirrors.aliyun.com/ubuntu|g' /etc/apt/sources.list && \
diff --git a/docker/inference/Dockerfile.vllm-amd b/docker/inference/Dockerfile.vllm-amd
index 0467a596f..52eb4a02c 100644
--- a/docker/inference/Dockerfile.vllm-amd
+++ b/docker/inference/Dockerfile.vllm-amd
@@ -1,4 +1,4 @@
-FROM vllm/vllm-openai-rocm:v0.19.0
+FROM vllm/vllm-openai-rocm:v0.20.1
 RUN pip config set global.index-url https://mirrors.aliyun.com/pypi/simple
 RUN pip install --no-cache-dir csghub-sdk==0.7.3 ray supervisor
 RUN sed -i 's|http://archive.ubuntu.com/ubuntu|http://mirrors.aliyun.com/ubuntu|g' /etc/apt/sources.list && \
diff --git a/docker/inference/README.md b/docker/inference/README.md
index 51901e6cc..0193a1ef8 100644
--- a/docker/inference/README.md
+++ b/docker/inference/README.md
@@ -12,15 +12,15 @@ echo "$OPENCSG_ACR_PASSWORD" | docker login $OPENCSG_ACR -u $OPENCSG_ACR_USERNAM
 ```bash
 export BUILDX_NO_DEFAULT_ATTESTATIONS=1
 
-# For vllm: opencsg-registry.cn-beijing.cr.aliyuncs.com/opencsghq/vllm:v0.8.5
-export IMAGE_TAG=v0.17.0
+# For vllm: opencsg-registry.cn-beijing.cr.aliyuncs.com/opencsghq/vllm:v0.20.1
+export IMAGE_TAG=v0.20.1
 docker buildx build --platform linux/amd64,linux/arm64 \
   -t ${OPENCSG_ACR}/opencsghq/vllm:${IMAGE_TAG} \
   -f Dockerfile.vllm \
   --push .
 
-# For amd-vllm: opencsg-registry.cn-beijing.cr.aliyuncs.com/opencsghq/amd-vllm:rocm7.0.0_vllm_0.19.0
-export IMAGE_TAG=rocm7.0.0_vllm_0.19.0
+# For amd-vllm: opencsg-registry.cn-beijing.cr.aliyuncs.com/opencsghq/amd-vllm:rocm7.2.1_vllm_0.20.1
+export IMAGE_TAG=rocm7.2.1_vllm_0.20.1
 docker buildx build --platform linux/amd64 \
   -t ${OPENCSG_ACR}/opencsghq/amd-vllm:${IMAGE_TAG} \
   -f Dockerfile.vllm-amd \