[0.34.0-dlc] CVE patching and use official vLLM wheel (#2911)

Author: ethnzhng

serving/docker/lmi-container-requirements.txt

@@ -1,7 +1,7 @@
 torch==2.8.0
 torchvision
 peft==0.15.1
-protobuf==3.20.3
+protobuf==4.25.1
 transformers==4.55.2
 hf-transfer
 zstandard
@@ -20,7 +20,7 @@ retrying
 opencv-contrib-python-headless
 safetensors
 scipy==1.16.0
-onnx==1.16.0
+onnx==1.19.0
 sentence_transformers
 onnxruntime-gpu==1.20.0
 autoawq
@@ -31,6 +31,6 @@ uvloop
 ninja
 peft
 llmcompressor
-https://publish.djl.ai/sm-vllm/vllm-0.10.2-cp38-abi3-linux_x86_64.whl
-xgrammar==0.1.23
+vllm==0.10.2
+xgrammar
 flashinfer-python==0.2.5

serving/docker/lmi.Dockerfile

@@ -89,8 +89,7 @@ RUN scripts/patch_oss_dlc.sh python \
     && apt-get clean -y && rm -rf /var/lib/apt/lists/*
 
 COPY lmi-container-requirements.txt ./requirements.txt
-RUN pip3 install torch==2.6.0 torchvision \
-    && pip3 install flashinfer-python -i https://flashinfer.ai/whl/cu126/torch2.6/ \
+RUN pip3 install torch==2.8.0 torchvision \
     && pip3 install -r requirements.txt \
     && pip3 install ${djl_converter_wheel} --no-deps