Release b8 (#301)

* fix

* enable b8

* Add

* delete

* fix

---------

Co-authored-by: liu-shaojun <shaojun.liu@intel.com>

Author: gc-fu

vllm/README.md

@@ -1193,6 +1193,18 @@ export VLLM_OFFLOAD_WEIGHTS_BEFORE_QUANT=1
 
 ## 5. Performance tuning
 
+### 5.1 Avoid Memory Fragmentation
+
+To avoid GPU memory fragmentation (which can lead to out-of-memory errors even when sufficient memory appears available), enable PyTorch's expandable segments feature:
+
+```bash
+export PYTORCH_ALLOC_CONF="expandable_segments:True"
+```
+
+Set this environment variable **before** launching the vLLM service. This allows PyTorch's memory allocator to use expandable segments instead of fixed-size blocks, significantly reducing fragmentation over long-running sessions.
+
+### 5.2 CPU Affinity (NUMA Binding)
+
 To improve performance, you can optimize CPU affinity based on the GPU–NUMA topology.
 
 For example, if your process uses two GPUs that are both connected to NUMA node 0, you can use lscpu to identify the CPU cores associated with that NUMA node:

vllm/docker/Dockerfile

@@ -2,7 +2,7 @@
 # SPDX-License-Identifier: Apache-2.0
 
 # ======== Base Stage ========
-FROM intel/llm-scaler-platform:26.5.6.1 AS vllm-base
+FROM intel/llm-scaler-platform:26.13.7.1 AS vllm-base
 
 ARG https_proxy
 ARG http_proxy
@@ -54,7 +54,7 @@ RUN python3 -m pip config set global.break-system-packages true
 
 # Clone + patch vllm
 RUN --mount=type=cache,target=/root/.cache/pip \
-    git clone -b v0.11.1 https://github.com/vllm-project/vllm.git && \
+    git clone -b v0.14.0 https://github.com/vllm-project/vllm.git && \
     cd vllm && \
     git apply /tmp/vllm_for_multi_arc.patch && \
     pip install -r requirements/xpu.txt && \
@@ -100,9 +100,14 @@ RUN --mount=type=cache,target=/root/.cache/pip \
 
 # Pin transformers version to avoid conflict in vLLM
 RUN --mount=type=cache,target=/root/.cache/pip \
-    pip install "transformers==4.57.3" && \
+    # pip install "transformers==4.57.3" && \
     pip install librosa soundfile decord
 
+# FIX triton
+RUN --mount=type=cache,target=/root/.cache/pip \
+    pip uninstall triton triton-xpu -y && \
+    pip install triton-xpu==3.6.0 --extra-index-url=https://download.pytorch.org/whl/test/xpu
+
 
 # Set additional environment for production usage
 ENV VLLM_QUANTIZE_Q40_LIB="/usr/local/lib/python3.12/dist-packages/vllm_int4_for_multi_arc.so"
@@ -116,26 +121,9 @@ RUN cd /llm/vllm && \
 
 RUN pip uninstall oneccl oneccl-devel -y
 
-ENV TBBROOT=/opt/intel/oneapi/tbb/2022.2/env/.. \
-    CCL_ROOT=/opt/intel/oneapi/ccl/2021.15.7-down.1 \
-    CMPLR_ROOT=/opt/intel/oneapi/compiler/2025.2 \
-    MKLROOT=/opt/intel/oneapi/mkl/2025.2 \
-    DPL_ROOT=/opt/intel/oneapi/dpl/2022.9 \
-    DNNLROOT=/opt/intel/oneapi/dnnl/2025.2 \
-    I_MPI_ROOT=/opt/intel/oneapi/mpi/2021.16
-
-
-ENV PKG_CONFIG_PATH=/opt/intel/oneapi/tbb/2022.2/env/../lib/pkgconfig:/opt/intel/oneapi/mpi/2021.16/lib/pkgconfig:/opt/intel/oneapi/mkl/2025.2/lib/pkgconfig:/opt/intel/oneapi/dpl/2022.9/lib/pkgconfig:/opt/intel/oneapi/dnnl/2025.2/lib/pkgconfig:/opt/intel/oneapi/compiler/2025.2/lib/pkgconfig:/opt/intel/oneapi/ccl/2021.15.7-down.1/lib/pkgconfig/
-
-ENV CMAKE_PREFIX_PATH=/opt/intel/oneapi/tbb/2022.2/env/..:/opt/intel/oneapi/pti/0.13/lib/cmake/pti:/opt/intel/oneapi/mkl/2025.2/lib/cmake:/opt/intel/oneapi/dpl/2022.9/lib/cmake/oneDPL:/opt/intel/oneapi/dnnl/2025.2/lib/cmake:/opt/intel/oneapi/compiler/2025.2:/opt/intel/oneapi/ccl/2021.15.7-down.1/lib/cmake/oneCCL
-
-ENV LIBRARY_PATH=/opt/intel/oneapi/tcm/1.4/lib:/opt/intel/oneapi/umf/0.11/lib:/opt/intel/oneapi/tbb/2022.2/env/../lib/intel64/gcc4.8:/opt/intel/oneapi/pti/0.13/lib:/opt/intel/oneapi/mpi/2021.16/lib:/opt/intel/oneapi/mkl/2025.2/lib:/opt/intel/oneapi/dnnl/2025.2/lib:/opt/intel/oneapi/compiler/2025.2/lib:/opt/intel/oneapi/ccl/2021.15.7-down.1/lib
-
-ENV LD_LIBRARY_PATH=/opt/intel/oneapi/tcm/1.4/lib:/opt/intel/oneapi/umf/0.11/lib:/opt/intel/oneapi/tbb/2022.2/env/../lib/intel64/gcc4.8:/opt/intel/oneapi/pti/0.13/lib:/opt/intel/oneapi/mpi/2021.16/opt/mpi/libfabric/lib:/opt/intel/oneapi/mpi/2021.16/lib:/opt/intel/oneapi/mkl/2025.2/lib:/opt/intel/oneapi/dnnl/2025.2/lib:/opt/intel/oneapi/debugger/2025.2/opt/debugger/lib:/opt/intel/oneapi/compiler/2025.2/opt/compiler/lib:/opt/intel/oneapi/compiler/2025.2/lib:/opt/intel/oneapi/ccl/2021.15.7-down.1/lib:/usr/local/lib/
+RUN rm /usr/lib/python3/dist-packages/PyJWT-2.7.0.dist-info/ -rf
 
-ENV CPLUS_INCLUDE_PATH=/opt/intel/oneapi/umf/0.11/include:/opt/intel/oneapi/tbb/2022.2/env/../include:/opt/intel/oneapi/pti/0.13/include:/opt/intel/oneapi/mpi/2021.16/include:/opt/intel/oneapi/mkl/2025.2/include:/opt/intel/oneapi/dpl/2022.9/include:/opt/intel/oneapi/dpcpp-ct/2025.2/include
-ENV CPATH=/opt/intel/oneapi/umf/0.11/include:/opt/intel/oneapi/mkl/2025.2/include:/opt/intel/oneapi/dnnl/2025.2/include:/opt/intel/oneapi/dev-utilities/2025.2/include:/opt/intel/oneapi/ccl/2021.15.7-down.1/include
+RUN echo "source /opt/intel/oneapi/setvars.sh --force" >> /root/.bashrc
 
-# ENTRYPOINT ["bash", "-c", "source /opt/intel/oneapi/setvars.sh --force && python3 -m vllm.entrypoints.openai.api_server"]
 ENTRYPOINT ["bash", "-c", "vllm serve"]