NVIDIA-NeMo · gwarmstrong · Dec 5, 2025 · Dec 5, 2025 · Dec 5, 2025
diff --git a/dockerfiles/Dockerfile.vllm b/dockerfiles/Dockerfile.vllm
@@ -0,0 +1,2 @@
+FROM vllm/vllm-openai:v0.10.2
+RUN pip install "vllm[audio]"
diff --git a/dockerfiles/README.md b/dockerfiles/README.md
@@ -36,4 +36,4 @@ We directly use official `lmsysorg/sglang:v0.5.4` image.
 
 ## Building vllm image
 
-We directly use official `vllm/vllm-openai:v0.10.2` image.
+We use official `vllm/vllm-openai:v0.10.2` image with the additional `vllm[audio]` dependencies.
diff --git a/nemo_skills/__init__.py b/nemo_skills/__init__.py
@@ -17,7 +17,7 @@
 # only used in ns setup command to initialize with defaults
 _containers = {
     "trtllm": "nvcr.io/nvidia/tensorrt-llm/release:1.0.0",
-    "vllm": "vllm/vllm-openai:v0.10.2",
+    "vllm": "dockerfile:dockerfiles/Dockerfile.vllm",
     "sglang": "lmsysorg/sglang:v0.5.4",
     "megatron": "dockerfile:dockerfiles/Dockerfile.megatron",
     "sandbox": "dockerfile:dockerfiles/Dockerfile.sandbox",

diff --git a/tests/gpu-tests/test-local.yaml b/tests/gpu-tests/test-local.yaml
@@ -16,7 +16,7 @@ executor: local
 
 containers:
   trtllm: nvcr.io/nvidia/tensorrt-llm/release:1.0.0
-  vllm: vllm/vllm-openai:v0.10.1.1
+  vllm: dockerfile:dockerfiles/Dockerfile.vllm
   sglang: lmsysorg/sglang:v0.5.4
   sandbox: dockerfile:dockerfiles/Dockerfile.sandbox
   nemo-skills: dockerfile:dockerfiles/Dockerfile.nemo-skills
Original file line number	Diff line number	Diff line change
Expand Up		@@ -36,4 +36,4 @@ We directly use official `lmsysorg/sglang:v0.5.4` image.

		## Building vllm image

		We directly use official `vllm/vllm-openai:v0.10.2` image.
		We use official `vllm/vllm-openai:v0.10.2` image with the additional `vllm[audio]` dependencies.