fix multiproc executor determine kv cache memory & update Dockerfile

Dockerfile.a3

@@ -22,11 +22,6 @@ ARG MOONCAKE_TAG=v0.3.8.post1
 ARG SOC_VERSION="ascend910_9391"
 
 COPY . /vllm-workspace/vllm-ascend/
-# Define environments
-ENV DEBIAN_FRONTEND=noninteractive
-ENV SOC_VERSION=$SOC_VERSION \
-    TASK_QUEUE_ENABLE=1 \
-    OMP_NUM_THREADS=1
 
 RUN pip config set global.index-url ${PIP_INDEX_URL}
 
@@ -57,6 +52,12 @@ RUN VLLM_TARGET_DEVICE="empty" python3 -m pip install -v -e /vllm-workspace/vllm
     python3 -m pip cache purge
 
 # Install vllm-ascend
+ENV DEBIAN_FRONTEND=noninteractive
+ENV SOC_VERSION=$SOC_VERSION \
+    TASK_QUEUE_ENABLE=1 \
+    OMP_NUM_THREADS=1 \
+    VLLM_WORKER_MULTIPROC_METHOD=spawn \
+    VLLM_ASCEND_ENABLE_VNPU=1
 # Append `libascend_hal.so` path (devlib) to LD_LIBRARY_PATH
 # Installing vllm-ascend on x86 can pull upstream triton back in alongside triton-ascend. Remove it immediately after this step.
 RUN export PIP_EXTRA_INDEX_URL=https://mirrors.huaweicloud.com/ascend/repos/pypi && \
@@ -85,9 +86,6 @@ RUN export PIP_EXTRA_INDEX_URL=https://mirrors.huaweicloud.com/ascend/repos/pypi
     python3 -m pip install "https://vllm-ascend.obs.cn-north-4.myhuaweicloud.com/vllm-ascend/${PTA_WHEEL}" && \
     python3 -m pip cache purge
 
-ENV VLLM_WORKER_MULTIPROC_METHOD=spawn \
-    VLLM_ASCEND_ENABLE_VNPU=1
-
 # Install clang-15 (for triton-ascend)
 RUN apt-get update -y && \
     apt-get -y install clang-15 && \