Merge branch 'main' into bump-version

Author: lvhan028

docker/Dockerfile

@@ -66,6 +66,5 @@ RUN --mount=type=cache,target=/root/.cache \
 # explicitly set ptxas path for triton
 ENV PATH=/opt/py3/bin:$PATH
 ENV TRITON_PTXAS_PATH=/usr/local/cuda/bin/ptxas
-ENV NCCL_LAUNCH_MODE=GROUP
 ENV LD_LIBRARY_PATH=/nccl/lib:$LD_LIBRARY_PATH
 FROM ${IMAGE_TYPE}

docker/prepare_wheel.sh

@@ -26,7 +26,7 @@ if [[ "${CUDA_VERSION_SHORT}" != "cu118" ]]; then
         DEEP_GEMM_VERSION=03d0be3
         FLASH_MLA_VERSION=9edee0c
     else
-        DEEP_GEMM_VERSION=1876566
+        DEEP_GEMM_VERSION=79f48ee
         FLASH_MLA_VERSION=c759027
     fi
 

k8s/deployment.yaml

@@ -24,8 +24,6 @@ spec:
         args:
         - "lmdeploy serve api_server internlm/internlm2-chat-7b --server-port 23333"
         env:
-        - name: NCCL_LAUNCH_MODE
-          value: GROUP
         - name: HUGGING_FACE_HUB_TOKEN
           value: "{{HUGGING_FACE_HUB_TOKEN}}"
         ports: