diff --git a/.github/Dockerfile.buildwheel b/.github/Dockerfile.buildwheel
index 960e397434f..5326935a874 100644
--- a/.github/Dockerfile.buildwheel
+++ b/.github/Dockerfile.buildwheel
@@ -15,7 +15,7 @@
 # This file is a part of the vllm-ascend project.
 #
 ARG PY_VERSION=3.11
-FROM quay.io/ascend/manylinux:8.2.rc1-910b-manylinux_2_28-py${PY_VERSION}
+FROM quay.io/ascend/manylinux:8.3.rc1-910b-manylinux_2_28-py${PY_VERSION}
 
 ARG COMPILE_CUSTOM_KERNELS=1
 
diff --git a/.github/workflows/_accuracy_test.yaml b/.github/workflows/_accuracy_test.yaml
index da02559e186..b9d155f2319 100644
--- a/.github/workflows/_accuracy_test.yaml
+++ b/.github/workflows/_accuracy_test.yaml
@@ -30,7 +30,7 @@ jobs:
     runs-on: ${{ inputs.runner }}
     name: ${{ inputs.model_name }} accuracy
     container:
-      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.2.rc1-910b-ubuntu22.04-py3.11
+      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.3.rc1-910b-ubuntu22.04-py3.11
       env:
         VLLM_USE_MODELSCOPE: True
         # 1. If version specified (work_dispatch), do specified branch accuracy test
diff --git a/.github/workflows/_e2e_nightly_multi_node.yaml b/.github/workflows/_e2e_nightly_multi_node.yaml
index db1a6414b85..3fa400e8a92 100644
--- a/.github/workflows/_e2e_nightly_multi_node.yaml
+++ b/.github/workflows/_e2e_nightly_multi_node.yaml
@@ -15,7 +15,7 @@ on:
         required: false
         type: string
         description: base image for pods
-        default: "swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.2.rc1-910b-ubuntu22.04-py3.11"
+        default: "swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.3.rc1-910b-ubuntu22.04-py3.11"
       config_file_path:
         required: true
         type: string
@@ -69,7 +69,7 @@ jobs:
     # This is the runner with no NPU for k8s controller
     runs-on: ${{ inputs.runner }}
     container:
-      image: m.daocloud.io/quay.io/ascend/cann:8.2.rc1-a3-ubuntu22.04-py3.11
+      image: m.daocloud.io/quay.io/ascend/cann:8.3.rc1-a3-ubuntu22.04-py3.11
       env:
         KUBECONFIG: /tmp/kubeconfig
         KUBECTL: /root/.cache/.kube/kubectl
diff --git a/.github/workflows/_e2e_nightly_single_node.yaml b/.github/workflows/_e2e_nightly_single_node.yaml
index e77f5623328..ab77649f429 100644
--- a/.github/workflows/_e2e_nightly_single_node.yaml
+++ b/.github/workflows/_e2e_nightly_single_node.yaml
@@ -29,7 +29,7 @@ on:
       image:
         required: false
         type: string
-        default: "swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.2.rc1-910b-ubuntu22.04-py3.11"
+        default: "swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.3.rc1-910b-ubuntu22.04-py3.11"
       tests:
         required: true
         type: string
diff --git a/.github/workflows/_kill_lws_resources.yaml b/.github/workflows/_kill_lws_resources.yaml
index dd6549bde52..4c70ff2d7d2 100644
--- a/.github/workflows/_kill_lws_resources.yaml
+++ b/.github/workflows/_kill_lws_resources.yaml
@@ -24,7 +24,7 @@ jobs:
     # This is a runner with no NPU for k8s controller
     runs-on: ${{ inputs.runner }}
     container:
-      image: m.daocloud.io/quay.io/ascend/cann:8.2.rc1-a3-ubuntu22.04-py3.11
+      image: m.daocloud.io/quay.io/ascend/cann:8.3.rc1-a3-ubuntu22.04-py3.11
       env:
         KUBECONFIG: /tmp/kubeconfig
         KUBECTL: /root/.cache/.kube/kubectl
diff --git a/.github/workflows/accuracy_test.yaml b/.github/workflows/accuracy_test.yaml
index a7d3c786b80..695cf70e48e 100644
--- a/.github/workflows/accuracy_test.yaml
+++ b/.github/workflows/accuracy_test.yaml
@@ -70,5 +70,5 @@ jobs:
     with:
       vllm: v0.11.0
       runner:  linux-aarch64-${{ matrix.runner }}
-      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.2.rc1-910b-ubuntu22.04-py3.11
+      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.3.rc1-910b-ubuntu22.04-py3.11
       model_name: ${{ matrix.model_name }}
diff --git a/.github/workflows/nightly_benchmarks.yaml b/.github/workflows/nightly_benchmarks.yaml
index 580e6333e62..a768cd66f5b 100644
--- a/.github/workflows/nightly_benchmarks.yaml
+++ b/.github/workflows/nightly_benchmarks.yaml
@@ -56,7 +56,7 @@ jobs:
             vllm_use_v1: 1
       max-parallel: 1
     container:
-      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.2.rc1-910b-ubuntu22.04-py3.11
+      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.3.rc1-910b-ubuntu22.04-py3.11
       volumes:
         - /usr/local/dcmi:/usr/local/dcmi
         - /usr/local/bin/npu-smi:/usr/local/bin/npu-smi
diff --git a/.github/workflows/vllm_ascend_dist.yaml b/.github/workflows/vllm_ascend_dist.yaml
index f78486e19e8..2d7c8ef235f 100644
--- a/.github/workflows/vllm_ascend_dist.yaml
+++ b/.github/workflows/vllm_ascend_dist.yaml
@@ -75,7 +75,7 @@ jobs:
     name: vLLM Ascend test
     runs-on: ${{ matrix.os }}
     container:
-      image: m.daocloud.io/quay.io/ascend/cann:8.2.rc1-a3-ubuntu22.04-py3.11
+      image: m.daocloud.io/quay.io/ascend/cann:8.3.rc1-a3-ubuntu22.04-py3.11
       env:
         DEBIAN_FRONTEND: noninteractive
     steps:
diff --git a/.github/workflows/vllm_ascend_test.yaml b/.github/workflows/vllm_ascend_test.yaml
index 23fb6ea6517..619e87158af 100644
--- a/.github/workflows/vllm_ascend_test.yaml
+++ b/.github/workflows/vllm_ascend_test.yaml
@@ -76,6 +76,7 @@ jobs:
     if: ${{ needs.lint.result == 'success' && (needs.changes.outputs.e2e_tracker == 'true' || needs.changes.outputs.ut_tracker == 'true') }}
     runs-on: ubuntu-latest
     container:
+      # fixme: vllm-ascend install failed with 8.3.rc1 on github action
       image: quay.io/ascend/cann:8.2.rc1-910b-ubuntu22.04-py3.11
       env:
         VLLM_LOGGING_LEVEL: ERROR
@@ -146,5 +147,5 @@ jobs:
     with:
       vllm: ${{ matrix.vllm_version }}
       runner: linux-aarch64-a2
-      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.2.rc1-910b-ubuntu22.04-py3.11
+      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.3.rc1-910b-ubuntu22.04-py3.11
       type: light
diff --git a/.github/workflows/vllm_ascend_test_310p.yaml b/.github/workflows/vllm_ascend_test_310p.yaml
index 1de447fc31d..099f3e07fe0 100644
--- a/.github/workflows/vllm_ascend_test_310p.yaml
+++ b/.github/workflows/vllm_ascend_test_310p.yaml
@@ -58,7 +58,7 @@ jobs:
     runs-on: ${{ matrix.os }}
     container:
       # TODO(yikun): Remove m.daocloud.io prefix when infra proxy ready
-      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.2.rc1-310p-ubuntu22.04-py3.11
+      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.3.rc1-310p-ubuntu22.04-py3.11
       env:
         VLLM_LOGGING_LEVEL: ERROR
         VLLM_USE_MODELSCOPE: True
diff --git a/.github/workflows/vllm_ascend_test_full.yaml b/.github/workflows/vllm_ascend_test_full.yaml
index 088036e6d8a..e16b7619670 100644
--- a/.github/workflows/vllm_ascend_test_full.yaml
+++ b/.github/workflows/vllm_ascend_test_full.yaml
@@ -76,5 +76,5 @@ jobs:
     with:
       vllm: ${{ matrix.vllm_version }}
       runner: linux-aarch64-a2
-      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.2.rc1-910b-ubuntu22.04-py3.11
+      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.3.rc1-910b-ubuntu22.04-py3.11
       type: full
diff --git a/.github/workflows/vllm_ascend_test_full_vllm_main.yaml b/.github/workflows/vllm_ascend_test_full_vllm_main.yaml
index 48dc695f8ca..dbd632912af 100644
--- a/.github/workflows/vllm_ascend_test_full_vllm_main.yaml
+++ b/.github/workflows/vllm_ascend_test_full_vllm_main.yaml
@@ -41,5 +41,5 @@ jobs:
     with:
       vllm: main
       runner: linux-aarch64-a2
-      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.2.rc1-910b-ubuntu22.04-py3.11
+      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.3.rc1-910b-ubuntu22.04-py3.11
       type: full
diff --git a/.github/workflows/vllm_ascend_test_models.yaml b/.github/workflows/vllm_ascend_test_models.yaml
index a26c5a3e232..beba0e44649 100644
--- a/.github/workflows/vllm_ascend_test_models.yaml
+++ b/.github/workflows/vllm_ascend_test_models.yaml
@@ -79,7 +79,7 @@ jobs:
     with:
       vllm: v0.11.0
       runner:  linux-aarch64-${{ matrix.runner }}
-      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.2.rc1-910b-ubuntu22.04-py3.11
+      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.3.rc1-910b-ubuntu22.04-py3.11
       model_name: ${{ matrix.model_name }}
       upload: ${{ github.event_name == 'workflow_dispatch' && github.event.inputs.vllm-ascend-version == 'latest' }}
 
diff --git a/.github/workflows/vllm_ascend_test_nightly_a2.yaml b/.github/workflows/vllm_ascend_test_nightly_a2.yaml
index 0842e7e10f1..19fc3b5dc9c 100644
--- a/.github/workflows/vllm_ascend_test_nightly_a2.yaml
+++ b/.github/workflows/vllm_ascend_test_nightly_a2.yaml
@@ -82,7 +82,7 @@ jobs:
     with:
       soc_version: a2
       runner: linux-aarch64-a2-0
-      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.2.rc1-910b-ubuntu22.04-py3.11
+      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.3.rc1-910b-ubuntu22.04-py3.11
       replicas: 1
       size: ${{ matrix.test_config.size }}
       config_file_path: ${{ matrix.test_config.config_file_path }}
diff --git a/.github/workflows/vllm_ascend_test_nightly_a3.yaml b/.github/workflows/vllm_ascend_test_nightly_a3.yaml
index 2cd6d817e9f..d880a8bffbe 100644
--- a/.github/workflows/vllm_ascend_test_nightly_a3.yaml
+++ b/.github/workflows/vllm_ascend_test_nightly_a3.yaml
@@ -82,7 +82,7 @@ jobs:
     with:
       vllm: v0.11.0
       runner: ${{ matrix.test_config.os }}
-      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.2.rc1-a3-ubuntu22.04-py3.11
+      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.3.rc1-a3-ubuntu22.04-py3.11
       tests: ${{ matrix.test_config.tests }}
 
   multi-node-tests:
@@ -113,7 +113,7 @@ jobs:
     with:
       soc_version: a3
       runner: linux-aarch64-a3-0
-      image: m.daocloud.io/quay.io/ascend/cann:8.2.rc1-a3-ubuntu22.04-py3.11
+      image: m.daocloud.io/quay.io/ascend/cann:8.3.rc1-a3-ubuntu22.04-py3.11
       replicas: 1
       size: ${{ matrix.test_config.size }}
       config_file_path: ${{ matrix.test_config.config_file_path }}
diff --git a/.github/workflows/vllm_ascend_test_pd.yaml b/.github/workflows/vllm_ascend_test_pd.yaml
index fee06bee829..778d83b6375 100644
--- a/.github/workflows/vllm_ascend_test_pd.yaml
+++ b/.github/workflows/vllm_ascend_test_pd.yaml
@@ -49,7 +49,7 @@ jobs:
     runs-on: linux-arm64-npu-static-8
 
     container:
-      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.2.rc1-910b-ubuntu22.04-py3.11
+      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.3.rc1-910b-ubuntu22.04-py3.11
       volumes:
         - /usr/local/dcmi:/usr/local/dcmi
         - /usr/local/bin/npu-smi:/usr/local/bin/npu-smi
diff --git a/Dockerfile b/Dockerfile
index e988028ddc4..f01284b4911 100644
--- a/Dockerfile
+++ b/Dockerfile
@@ -15,7 +15,7 @@
 # This file is a part of the vllm-ascend project.
 #
 
-FROM quay.io/ascend/cann:8.2.rc1-910b-ubuntu22.04-py3.11
+FROM quay.io/ascend/cann:8.3.rc1-910b-ubuntu22.04-py3.11
 
 ARG PIP_INDEX_URL="https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple"
 ARG COMPILE_CUSTOM_KERNELS=1
diff --git a/Dockerfile.310p b/Dockerfile.310p
index 212058954c4..423f2143baa 100644
--- a/Dockerfile.310p
+++ b/Dockerfile.310p
@@ -15,7 +15,7 @@
 # This file is a part of the vllm-ascend project.
 #
 
-FROM quay.io/ascend/cann:8.2.rc1-310p-ubuntu22.04-py3.11
+FROM quay.io/ascend/cann:8.3.rc1-310p-ubuntu22.04-py3.11
 
 ARG PIP_INDEX_URL="https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple"
 ARG COMPILE_CUSTOM_KERNELS=1
diff --git a/Dockerfile.310p.openEuler b/Dockerfile.310p.openEuler
index 38015239871..4da46b10602 100644
--- a/Dockerfile.310p.openEuler
+++ b/Dockerfile.310p.openEuler
@@ -15,7 +15,7 @@
 # This file is a part of the vllm-ascend project.
 #
 
-FROM quay.io/ascend/cann:8.2.rc1-310p-openeuler24.03-py3.11
+FROM quay.io/ascend/cann:8.3.rc1-310p-openeuler24.03-py3.11
 
 ARG PIP_INDEX_URL="https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple"
 ARG COMPILE_CUSTOM_KERNELS=1
diff --git a/Dockerfile.a3 b/Dockerfile.a3
index 320e8fc75c2..59c4cbbbebd 100644
--- a/Dockerfile.a3
+++ b/Dockerfile.a3
@@ -15,7 +15,7 @@
 # This file is a part of the vllm-ascend project.
 #
 
-FROM quay.io/ascend/cann:8.2.rc1-a3-ubuntu22.04-py3.11
+FROM quay.io/ascend/cann:8.3.rc1-a3-ubuntu22.04-py3.11
 
 ARG PIP_INDEX_URL="https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple"
 ARG COMPILE_CUSTOM_KERNELS=1
diff --git a/Dockerfile.a3.openEuler b/Dockerfile.a3.openEuler
index 1e6ba351cb8..2fabb3069a0 100644
--- a/Dockerfile.a3.openEuler
+++ b/Dockerfile.a3.openEuler
@@ -15,7 +15,7 @@
 # This file is a part of the vllm-ascend project.
 #
 
-FROM quay.io/ascend/cann:8.2.rc1-a3-openeuler24.03-py3.11
+FROM quay.io/ascend/cann:8.3.rc1-a3-openeuler24.03-py3.11
 
 ARG PIP_INDEX_URL="https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple"
 ARG COMPILE_CUSTOM_KERNELS=1
diff --git a/Dockerfile.openEuler b/Dockerfile.openEuler
index 2486dbae642..e665518d97d 100644
--- a/Dockerfile.openEuler
+++ b/Dockerfile.openEuler
@@ -15,7 +15,7 @@
 # This file is a part of the vllm-ascend project.
 #
 
-FROM quay.io/ascend/cann:8.2.rc1-910b-openeuler24.03-py3.11
+FROM quay.io/ascend/cann:8.3.rc1-910b-openeuler24.03-py3.11
 
 ARG PIP_INDEX_URL="https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple"
 ARG COMPILE_CUSTOM_KERNELS=1
diff --git a/README.md b/README.md
index b7952e54b30..1d0529bfd0f 100644
--- a/README.md
+++ b/README.md
@@ -42,7 +42,7 @@ By using vLLM Ascend plugin, popular open-source models, including Transformer-l
 - OS: Linux
 - Software:
   * Python >= 3.9, < 3.12
-  * CANN >= 8.2.rc1 (Ascend HDK version refers to [here](https://www.hiascend.com/document/detail/zh/canncommercial/82RC1/releasenote/releasenote_0000.html))
+  * CANN >= 8.3.rc1 (Ascend HDK version refers to [here](https://www.hiascend.com/document/detail/zh/canncommercial/83RC1/releasenote/releasenote_0000.html))
   * PyTorch == 2.7.1, torch-npu == 2.7.1
   * vLLM (the same version as vllm-ascend)
 
diff --git a/README.zh.md b/README.zh.md
index 1840de021df..a28056fa3db 100644
--- a/README.zh.md
+++ b/README.zh.md
@@ -43,7 +43,7 @@ vLLM 昇腾插件 (`vllm-ascend`) 是一个由社区维护的让vLLM在Ascend NP
 - 操作系统：Linux
 - 软件：
   * Python >= 3.9, < 3.12
-  * CANN >= 8.2.rc1 (Ascend HDK 版本参考[这里](https://www.hiascend.com/document/detail/zh/canncommercial/82RC1/releasenote/releasenote_0000.html))
+  * CANN >= 8.3.rc1 (Ascend HDK 版本参考[这里](https://www.hiascend.com/document/detail/zh/canncommercial/83RC1/releasenote/releasenote_0000.html))
   * PyTorch == 2.7.1, torch-npu == 2.7.1
   * vLLM (与vllm-ascend版本一致)
 
diff --git a/docs/source/conf.py b/docs/source/conf.py
index 766f1e2f282..149f327581d 100644
--- a/docs/source/conf.py
+++ b/docs/source/conf.py
@@ -75,7 +75,7 @@
     'pip_vllm_ascend_version': "0.11.0rc0",
     'pip_vllm_version': "0.11.0",
     # CANN image tag
-    'cann_image_tag': "8.2.rc1-910b-ubuntu22.04-py3.11",
+    'cann_image_tag': "8.3.rc1-910b-ubuntu22.04-py3.11",
     # vllm version in ci
     'ci_vllm_version': 'v0.11.0',
 }
diff --git a/docs/source/developer_guide/contribution/multi_node_test.md b/docs/source/developer_guide/contribution/multi_node_test.md
index 6e96084cca7..1d78c8e3536 100644
--- a/docs/source/developer_guide/contribution/multi_node_test.md
+++ b/docs/source/developer_guide/contribution/multi_node_test.md
@@ -90,7 +90,7 @@ currently, the multi-node test workflow defined in the [vllm_ascend_test_nightly
         uses: ./.github/workflows/_e2e_nightly_multi_node.yaml
         with:
         soc_version: a3
-        image: m.daocloud.io/quay.io/ascend/cann:8.2.rc1-a3-ubuntu22.04-py3.11
+        image: m.daocloud.io/quay.io/ascend/cann:8.3.rc1-a3-ubuntu22.04-py3.11
         replicas: 1
         size: ${{ matrix.test_config.size }}
         config_file_path: ${{ matrix.test_config.config_file_path }}
diff --git a/docs/source/installation.md b/docs/source/installation.md
index a08e9e22b35..e8bc8cbbeac 100644
--- a/docs/source/installation.md
+++ b/docs/source/installation.md
@@ -11,8 +11,8 @@ This document describes how to install vllm-ascend manually.
 
     | Software      | Supported version                | Note                                      |
     |---------------|----------------------------------|-------------------------------------------|
-    | Ascend HDK    | Refer to [here](https://www.hiascend.com/document/detail/zh/canncommercial/82RC1/releasenote/releasenote_0000.html) | Required for CANN |
-    | CANN          | >= 8.2.RC1                       | Required for vllm-ascend and torch-npu    |
+    | Ascend HDK    | Refer to [here](https://www.hiascend.com/document/detail/zh/canncommercial/83RC1/releasenote/releasenote_0000.html) | Required for CANN |
+    | CANN          | >= 8.3.RC1                       | Required for vllm-ascend and torch-npu    |
     | torch-npu     | == 2.7.1             | Required for vllm-ascend, No need to install manually, it will be auto installed in below steps |
     | torch         | == 2.7.1                         | Required for torch-npu and vllm           |
 
@@ -80,19 +80,19 @@ source vllm-ascend-env/bin/activate
 pip3 install -i https://pypi.tuna.tsinghua.edu.cn/simple attrs 'numpy<2.0.0' decorator sympy cffi pyyaml pathlib2 psutil protobuf scipy requests absl-py wheel typing_extensions
 
 # Download and install the CANN package.
-wget --header="Referer: https://www.hiascend.com/" https://ascend-repo.obs.cn-east-2.myhuaweicloud.com/CANN/CANN%208.2.RC1/Ascend-cann-toolkit_8.2.RC1_linux-"$(uname -i)".run
-chmod +x ./Ascend-cann-toolkit_8.2.RC1_linux-"$(uname -i)".run
-./Ascend-cann-toolkit_8.2.RC1_linux-"$(uname -i)".run --full
-# https://ascend-repo.obs.cn-east-2.myhuaweicloud.com/Milan-ASL/Milan-ASL%20V100R001C22B800TP052/Ascend-cann-kernels-910b_8.2.rc1_linux-aarch64.run
+wget --header="Referer: https://www.hiascend.com/" https://ascend-repo.obs.cn-east-2.myhuaweicloud.com/CANN/CANN%208.3.RC1/Ascend-cann-toolkit_8.3.RC1_linux-"$(uname -i)".run
+chmod +x ./Ascend-cann-toolkit_8.3.RC1_linux-"$(uname -i)".run
+./Ascend-cann-toolkit_8.3.RC1_linux-"$(uname -i)".run --full
+# https://ascend-repo.obs.cn-east-2.myhuaweicloud.com/Milan-ASL/Milan-ASL%20V100R001C22B800TP052/Ascend-cann-kernels-910b_8.3.rc1_linux-aarch64.run
 
 source /usr/local/Ascend/ascend-toolkit/set_env.sh
-wget --header="Referer: https://www.hiascend.com/" https://ascend-repo.obs.cn-east-2.myhuaweicloud.com/CANN/CANN%208.2.RC1/Ascend-cann-kernels-910b_8.2.RC1_linux-"$(uname -i)".run
-chmod +x ./Ascend-cann-kernels-910b_8.2.RC1_linux-"$(uname -i)".run
-./Ascend-cann-kernels-910b_8.2.RC1_linux-"$(uname -i)".run --install
+wget --header="Referer: https://www.hiascend.com/" https://ascend-repo.obs.cn-east-2.myhuaweicloud.com/CANN/CANN%208.3.RC1/Ascend-cann-kernels-910b_8.3.RC1_linux-"$(uname -i)".run
+chmod +x ./Ascend-cann-kernels-910b_8.3.RC1_linux-"$(uname -i)".run
+./Ascend-cann-kernels-910b_8.3.RC1_linux-"$(uname -i)".run --install
 
-wget --header="Referer: https://www.hiascend.com/" https://ascend-repo.obs.cn-east-2.myhuaweicloud.com/CANN/CANN%208.2.RC1/Ascend-cann-nnal_8.2.RC1_linux-"$(uname -i)".run
-chmod +x ./Ascend-cann-nnal_8.2.RC1_linux-"$(uname -i)".run
-./Ascend-cann-nnal_8.2.RC1_linux-"$(uname -i)".run --install
+wget --header="Referer: https://www.hiascend.com/" https://ascend-repo.obs.cn-east-2.myhuaweicloud.com/CANN/CANN%208.3.RC1/Ascend-cann-nnal_8.3.RC1_linux-"$(uname -i)".run
+chmod +x ./Ascend-cann-nnal_8.3.RC1_linux-"$(uname -i)".run
+./Ascend-cann-nnal_8.3.RC1_linux-"$(uname -i)".run --install
 
 source /usr/local/Ascend/nnal/atb/set_env.sh
 ```
diff --git a/examples/disaggregated_prefill_v1/mooncake_connector_deployment_guide.md b/examples/disaggregated_prefill_v1/mooncake_connector_deployment_guide.md
index ea76f0d511d..563357f8ba5 100644
--- a/examples/disaggregated_prefill_v1/mooncake_connector_deployment_guide.md
+++ b/examples/disaggregated_prefill_v1/mooncake_connector_deployment_guide.md
@@ -4,7 +4,7 @@
 
  *  Software:
      *  Python >= 3.9, < 3.12
-     *  CANN >= 8.2.rc1
+     *  CANN >= 8.3.rc1
      *  PyTorch == 2.7.1, torch-npu == 2.7.1
      *  vLLM (same version as vllm-ascend)
      *  mooncake-transfer-engine reference documentation: https://github.com/kvcache-ai/Mooncake/blob/main/doc/zh/ascend_transport.md
diff --git a/examples/disaggregated_prefill_v1/mooncake_connector_store_deployment_guide.md b/examples/disaggregated_prefill_v1/mooncake_connector_store_deployment_guide.md
index 82640215a84..28dd83b7cd0 100644
--- a/examples/disaggregated_prefill_v1/mooncake_connector_store_deployment_guide.md
+++ b/examples/disaggregated_prefill_v1/mooncake_connector_store_deployment_guide.md
@@ -4,7 +4,7 @@
 
 * Software:
   * Python >= 3.9, < 3.12
-  * CANN >= 8.2.rc1
+  * CANN >= 8.3.rc1
   * PyTorch == 2.7.1, torch-npu == 2.7.1
   * vLLM：main branch
   * vLLM-Ascend：main branch
diff --git a/tests/e2e/nightly/multi_node/scripts/lws.yaml.jinja2 b/tests/e2e/nightly/multi_node/scripts/lws.yaml.jinja2
index f619b59754d..712fe3f06b0 100644
--- a/tests/e2e/nightly/multi_node/scripts/lws.yaml.jinja2
+++ b/tests/e2e/nightly/multi_node/scripts/lws.yaml.jinja2
@@ -15,7 +15,7 @@ spec:
       spec:
         containers:
           - name: vllm-leader
-            image: {{ image | default("m.daocloud.io/quay.io/ascend/cann:8.2.rc1-a3-ubuntu22.04-py3.11") }}
+            image: {{ image | default("m.daocloud.io/quay.io/ascend/cann:8.3.rc1-a3-ubuntu22.04-py3.11") }}
             env:
               - name: CONFIG_YAML_PATH
                 value: {{ config_file_path | default("DeepSeek-V3.yaml") }}
@@ -75,7 +75,7 @@ spec:
       spec:
         containers:
           - name: vllm-worker
-            image: {{ image | default("m.daocloud.io/quay.io/ascend/cann:8.2.rc1-a3-ubuntu22.04-py3.11") }}
+            image: {{ image | default("m.daocloud.io/quay.io/ascend/cann:8.3.rc1-a3-ubuntu22.04-py3.11") }}
             env:
               - name: CONFIG_YAML_PATH
                 value: {{ config_file_path | default("DeepSeek-V3.yaml") }}
diff --git a/tests/e2e/vllm_interface/vllm_test.cfg b/tests/e2e/vllm_interface/vllm_test.cfg
index 4d077b09092..9723d49cad7 100644
--- a/tests/e2e/vllm_interface/vllm_test.cfg
+++ b/tests/e2e/vllm_interface/vllm_test.cfg
@@ -1,2 +1,2 @@
 # Base docker image used to build the vllm-ascend e2e test image, which is built in the vLLM repository
-BASE_IMAGE_NAME="quay.io/ascend/cann:8.2.rc1-910b-ubuntu22.04-py3.11"
+BASE_IMAGE_NAME="quay.io/ascend/cann:8.3.rc1-910b-ubuntu22.04-py3.11"
diff --git a/tests/ut/attention/test_attention_mask.py b/tests/ut/attention/test_attention_mask.py
index a87d21bd742..c8139b7167b 100644
--- a/tests/ut/attention/test_attention_mask.py
+++ b/tests/ut/attention/test_attention_mask.py
@@ -74,11 +74,10 @@ def test_get_attn_mask(self):
         attn_mask = attention_mask_builder.get_attn_mask(
             max_seq_len=2048, dtype=torch.float16, device=torch.device("cpu"))
         self.assertEqual(attn_mask.shape, (2048, 2048))
-        self.assertEqual(attn_mask[0][-1],
-                         torch.tensor(float("-inf"), dtype=torch.float16))
-        self.assertEqual(attention_mask_builder._seq_len_cached, 2048)
+        self.assertEqual(attn_mask[0][-1], torch.tensor(True))
+        self.assertEqual(attention_mask_builder._seq_len_cached, 1024)
         self.assertEqual(attention_mask_builder.attn_mask_cache.shape,
-                         (2048, 2048))
+                         (1024, 1024))
         self.assertEqual(attention_mask_builder.attn_mask_cache[0][-1],
                          torch.tensor(float("-inf"), dtype=torch.float16))
 
@@ -91,43 +90,5 @@ def test_get_splitfuse_attn_mask(self):
             dtype=torch.float16,
             device=torch.device("cpu"),
         )
-        self.assertEqual(attn_mask.shape, (6, 100))
+        self.assertEqual(attn_mask.shape, (2048, 2048))
         self.assertEqual(attention_mask_builder._seq_len_cached, 1024)
-
-        attn_mask = attention_mask_builder.get_splitfuse_attn_mask(
-            seq_lens=torch.tensor([10, 3000, 2000]),
-            position=torch.tensor([7, 8, 9, 2999, 1999]),
-            dtype=torch.float16,
-            device=torch.device("cpu"),
-        )
-        self.assertEqual(attn_mask.shape, (5, 3000))
-        self.assertEqual(attention_mask_builder._seq_len_cached, 3000)
-
-        # splitfuse_attn_mask now only supports data types: torch.float16 and torch.bfloat16
-        # otherwise raise ValueError
-        with self.assertRaises(ValueError):
-            attn_mask = attention_mask_builder.get_splitfuse_attn_mask(
-                seq_lens=torch.tensor([10, 20, 100]),
-                position=torch.tensor([7, 8, 9, 18, 19, 99]),
-                dtype=torch.int8,
-                device=torch.device("cpu"),
-            )
-
-    def test_mask_value_cleanliness(self):
-        attention_mask_builder = AttentionMaskBuilder(max_seq_len=6,
-                                                      dtype=torch.bfloat16)
-        self.assertEqual(attention_mask_builder.attn_mask_cache[-2][-1],
-                         torch.tensor(1, dtype=torch.bfloat16))
-
-        attn_mask = attention_mask_builder.get_splitfuse_attn_mask(
-            seq_lens=torch.tensor([6]),
-            position=torch.tensor([3, 4, 5]),
-            dtype=torch.bfloat16,
-            device=torch.device("cpu"),
-        )
-        self.assertEqual(
-            attn_mask[-2][-1],
-            torch.tensor(-10000, dtype=torch.bfloat16,
-                         device=attn_mask.device))
-        self.assertEqual(attention_mask_builder.attn_mask_cache[-2][-1],
-                         torch.tensor(1, dtype=torch.bfloat16))
diff --git a/tests/ut/attention/test_attention_v1.py b/tests/ut/attention/test_attention_v1.py
index 237fd299fd9..dfb9a2a07f8 100644
--- a/tests/ut/attention/test_attention_v1.py
+++ b/tests/ut/attention/test_attention_v1.py
@@ -298,8 +298,9 @@ def test_forward_prefill_no_cache(self, mock_flash_attention,
         assert output.shape == (10, 8 * 64)
 
     @patch('torch_npu._npu_reshape_and_cache')
-    @patch('torch_npu._npu_flash_attention_qlens')
-    def test_forward_prefill_cache_hit(self, mock_flash_attention_qlens,
+    @patch('torch_npu.npu_fused_infer_attention_score')
+    def test_forward_prefill_cache_hit(self,
+                                       mock_npu_fused_infer_attention_score,
                                        mock_npu_reshape_and_cache):
         """Test forward pass in PrefillCacheHit state"""
         query = torch.randn(10, 8 * 64)
@@ -308,6 +309,8 @@ def test_forward_prefill_cache_hit(self, mock_flash_attention_qlens,
         kv_cache = torch.empty(2, 5, 128, 8, 64)
         output = torch.empty_like(query)
 
+        mock_npu_fused_infer_attention_score.return_value = (output, 1)
+
         metadata = self.attn_metadata
         metadata.attn_state = AscendAttentionState.PrefillCacheHit
         metadata.attn_mask = torch.randn(1, 1, 10, 10)
@@ -323,7 +326,7 @@ def test_forward_prefill_cache_hit(self, mock_flash_attention_qlens,
         output = self.impl.forward(layer, query, key, value, kv_cache,
                                    metadata, output)
 
-        mock_flash_attention_qlens.assert_called_once()
+        mock_npu_fused_infer_attention_score.assert_called_once()
         assert output.shape == (10, 8 * 64)
 
     @patch('vllm_ascend.attention.attention_v1.get_forward_context')
@@ -528,13 +531,11 @@ def test_forward_decode_only_swa_seq_len_mismatch(
 
         assert output.shape == (10, 8 * 64)
 
-    @patch('torch.version')
     @patch('vllm_ascend.attention.attention_v1.is_310p', return_value=False)
     @patch('torch_npu._npu_reshape_and_cache')
     @patch('vllm_ascend.attention.attention_v1.vanilla_chunked_prefill')
     def test_forward_head_size_192(self, mock_vanilla_prefill,
-                                   mock_npu_reshape_and_cache, mock_is_310p,
-                                   mock_version):
+                                   mock_npu_reshape_and_cache, mock_is_310p):
         """Test forward pass when head_size is 192"""
 
         self.impl.head_size = 192
@@ -554,7 +555,6 @@ def test_forward_head_size_192(self, mock_vanilla_prefill,
         metadata.num_decodes = 10
         metadata.num_prefills = 0
         layer = self.layer_no_quant
-        mock_version.cann = "8.4.RC1"
         mock_vanilla_prefill.return_value = MagicMock()
 
         output = self.impl_192.forward(layer, query, key, value, kv_cache,
@@ -563,12 +563,11 @@ def test_forward_head_size_192(self, mock_vanilla_prefill,
         mock_vanilla_prefill.assert_called_once()
         assert output.shape == (10, 8 * 192)
 
-    @patch('torch.version')
     @patch('torch_npu._npu_reshape_and_cache')
-    @patch('torch_npu._npu_paged_attention_splitfuse')
-    def test_forward_normal_v1_situation(self, mock_paged_attention,
-                                         mock_npu_reshape_and_cache,
-                                         mock_version):
+    @patch('torch_npu.npu_fused_infer_attention_score')
+    def test_forward_normal_v1_situation(self,
+                                         mock_npu_fused_infer_attention_score,
+                                         mock_npu_reshape_and_cache):
         """Test forward pass in normal V1 situation"""
         query = torch.randn(10, 8 * 64)
         key = torch.randn(10, 8 * 64)
@@ -576,6 +575,8 @@ def test_forward_normal_v1_situation(self, mock_paged_attention,
         kv_cache = torch.empty(2, 5, 128, 8, 64)
         output = torch.empty_like(query)
 
+        mock_npu_fused_infer_attention_score.return_value = (output, 1)
+
         metadata = self.attn_metadata
         metadata.attn_mask = torch.randn(1, 1, 10, 10)
         metadata.query_lens = torch.tensor([10])
@@ -587,22 +588,20 @@ def test_forward_normal_v1_situation(self, mock_paged_attention,
         metadata.num_prefills = 10
         layer = self.layer_no_quant
 
-        mock_version.cann = "8.4.RC1"
-
         output = self.impl.forward(layer, query, key, value, kv_cache,
                                    metadata, output)
 
-        mock_paged_attention.assert_called_once()
+        mock_npu_fused_infer_attention_score.assert_called_once()
         assert output.shape == (10, 8 * 64)
 
-    @patch('torch.version')
     @patch('torch_npu.npu_format_cast')
     @patch('torch_npu._npu_reshape_and_cache')
-    @patch('torch_npu._npu_paged_attention_splitfuse')
+    @patch('torch_npu.npu_fused_infer_attention_score')
     @patch('vllm_ascend.attention.attention_v1.is_310p', return_value=True)
-    def test_forward_310p_device(self, mock_is_310p, mock_paged_attention,
+    def test_forward_310p_device(self, mock_is_310p,
+                                 mock_npu_fused_infer_attention_score,
                                  mock_npu_reshape_and_cache,
-                                 mock_npu_format_cast, mock_version):
+                                 mock_npu_format_cast):
         """Test forward pass on 310P device"""
         query = torch.randn(10, 8 * 64)
         key = torch.randn(10, 8 * 64)
@@ -610,6 +609,8 @@ def test_forward_310p_device(self, mock_is_310p, mock_paged_attention,
         kv_cache = torch.empty(2, 5, 128, 8, 64)
         output = torch.empty_like(query)
 
+        mock_npu_fused_infer_attention_score.return_value = (output, 1)
+
         metadata = self.attn_metadata
         metadata.attn_mask = torch.randn(1, 1, 10, 10)
         metadata.query_lens = torch.tensor([10])
@@ -622,12 +623,11 @@ def test_forward_310p_device(self, mock_is_310p, mock_paged_attention,
         layer = self.layer_no_quant
 
         mock_npu_format_cast.return_value = metadata.attn_mask
-        mock_version.cann = "8.4.RC1"
 
         output = self.impl.forward(layer, query, key, value, kv_cache,
                                    metadata, output)
 
-        mock_paged_attention.assert_called_once()
+        mock_npu_fused_infer_attention_score.assert_called_once()
         assert output.shape == (10, 8 * 64)
 
     @patch('torch_npu._npu_reshape_and_cache')
diff --git a/tests/ut/ops/test_linear.py b/tests/ut/ops/test_linear.py
index bc1751ac873..1b3a7268fc6 100644
--- a/tests/ut/ops/test_linear.py
+++ b/tests/ut/ops/test_linear.py
@@ -63,33 +63,20 @@ def setUp(self):
 
     @mock.patch("vllm_ascend.ops.linear.is_enable_nz")
     @mock.patch("torch_npu.npu_format_cast")
-    @mock.patch("torch.version")
-    def test_process_weights_after_loading_is_8_3_enable_nz(
-            self, mock_version, mock_format_cast, mock_is_nz):
-        mock_version.cann = "8.3.RC1"
+    def test_process_weights_after_loading_enable_nz(self, mock_format_cast,
+                                                     mock_is_nz):
         mock_is_nz.return_value = 1
         self.method.process_weights_after_loading(self.layer)
         mock_format_cast.assert_called_once()
 
     @mock.patch("vllm_ascend.ops.linear.is_enable_nz")
     @mock.patch("torch_npu.npu_format_cast")
-    @mock.patch("torch.version")
-    def test_process_weights_after_loading_is_8_3_disable_nz(
-            self, mock_version, mock_format_cast, mock_is_nz):
-        mock_version.cann = "8.3.RC1"
+    def test_process_weights_after_loading_disable_nz(self, mock_format_cast,
+                                                      mock_is_nz):
         mock_is_nz.return_value = 0
         self.method.process_weights_after_loading(self.layer)
         mock_format_cast.assert_not_called()
 
-    @mock.patch("vllm_ascend.ops.linear.is_enable_nz")
-    @mock.patch("torch.version")
-    def test_process_weights_after_loading_not_8_3(self, mock_version,
-                                                   mock_is_nz):
-        mock_version.cann = "8.2.RC1"
-        mock_is_nz.return_value = 1
-        # Should not raise exception
-        self.method.process_weights_after_loading(self.layer)
-
 
 class TestAscendRowParallelLinear(BaseLinearTest):
 
diff --git a/vllm_ascend/attention/attention_mask.py b/vllm_ascend/attention/attention_mask.py
index 5a94b82b149..3514984d826 100644
--- a/vllm_ascend/attention/attention_mask.py
+++ b/vllm_ascend/attention/attention_mask.py
@@ -47,11 +47,10 @@ def __init__(
         self.attn_mask_cache = attn_mask
         self.device = device
         self.pooling_mask = None
-        if torch.version.cann.startswith("8.3"):
-            assigned_mask_dim = 2048
-            self.chunked_prefill_attn_mask = torch.triu(
-                torch.ones(assigned_mask_dim, assigned_mask_dim),
-                diagonal=1).to(torch.int8).to(device)
+        assigned_mask_dim = 2048
+        self.chunked_prefill_attn_mask = torch.triu(
+            torch.ones(assigned_mask_dim, assigned_mask_dim),
+            diagonal=1).to(torch.int8).to(device)
 
     @staticmethod
     def get_mask_scale_factor(dtype: torch.dtype = torch.float16):
@@ -68,7 +67,7 @@ def get_mask_scale_factor(dtype: torch.dtype = torch.float16):
 
     def get_attn_mask(self, max_seq_len: int, dtype: torch.dtype,
                       device: torch.device):
-        if max_seq_len == 2048 and torch.version.cann.startswith("8.3"):
+        if max_seq_len == 2048:
             return self.chunked_prefill_attn_mask.to(torch.bool)
         self._update_attn_cache(max_seq_len, dtype)
         return self.attn_mask_cache[:max_seq_len, :max_seq_len].contiguous(
@@ -89,23 +88,7 @@ def get_splitfuse_attn_mask(
         dtype: torch.dtype = None,
         device: torch.device = None,
     ) -> torch.Tensor:
-        if torch.version.cann.startswith("8.3"):
-            return self.chunked_prefill_attn_mask
-        else:
-            if dtype not in [torch.float16, torch.bfloat16]:
-                raise ValueError(
-                    "splitfuse_attn_mask now only supports bf16 and fp16")
-            max_seq_len = max(seq_lens, default=0)
-            self._update_attn_cache(max_seq_len, dtype)
-            # FIXME: Currently the mask value of chunked-prefill situation and Prefill-Only situation
-            # is not the same. Fix this in the future when kernel is ready.
-            mask_scale_factor = AttentionMaskBuilder.get_mask_scale_factor(
-                dtype)
-            attn_mask = torch.index_select(self.attn_mask_cache,
-                                           dim=0,
-                                           index=position)[:, :max_seq_len]
-            attn_mask *= mask_scale_factor
-            return attn_mask.contiguous().to(device, non_blocking=True)
+        return self.chunked_prefill_attn_mask
 
     def _update_attn_cache(self, seqlen: int, dtype: torch.dtype):
         if seqlen > self._seq_len_cached:
diff --git a/vllm_ascend/attention/attention_v1.py b/vllm_ascend/attention/attention_v1.py
index fb29c9dd6e0..59d6fcb2c1e 100644
--- a/vllm_ascend/attention/attention_v1.py
+++ b/vllm_ascend/attention/attention_v1.py
@@ -500,7 +500,7 @@ def _forward_prefill_cache_hit(
         block_table = attn_metadata.block_tables[:batch_size, :]
         num_block, block_size, _, _ = self.key_cache.shape  # type: ignore
 
-        if torch.version.cann.startswith("8.3") and block_size == 128:
+        if block_size == 128:
             # TODO:The npu_fused_infer_attention_score op is planned to
             # be utilized in a wider range in upcoming versions.
             key = self.key_cache.view(  # type: ignore
@@ -680,43 +680,30 @@ def _forward_v1_style(
             attn_metadata.seq_lens = \
                 attn_metadata.seq_lens.to(device=query.device)
 
-        if torch.version.cann.startswith("8.3"):
-            # TODO:The npu_fused_infer_attention_score op is planned to
-            # be utilized in a wider range in upcoming versions.
-            num_block, block_size, _, _ = self.key_cache.shape  # type: ignore
-            key = self.key_cache.view(  # type: ignore
-                num_block, block_size, -1)
-            value = self.value_cache.view(  # type: ignore
-                num_block, block_size, -1)
+        # TODO:The npu_fused_infer_attention_score op is planned to
+        # be utilized in a wider range in upcoming versions.
+        num_block, block_size, _, _ = self.key_cache.shape  # type: ignore
+        key = self.key_cache.view(  # type: ignore
+            num_block, block_size, -1)
+        value = self.value_cache.view(  # type: ignore
+            num_block, block_size, -1)
+
+        output, _ = torch_npu.npu_fused_infer_attention_score(
+            query=query,
+            key=key,
+            value=value,
+            atten_mask=attn_metadata.attn_mask,
+            block_table=attn_metadata.block_tables,
+            input_layout="TND",
+            block_size=block_size,
+            actual_seq_lengths=attn_metadata.actual_seq_lengths_q,
+            actual_seq_lengths_kv=attn_metadata.seq_lens_list,
+            num_key_value_heads=self.num_kv_heads,
+            num_heads=self.num_heads,
+            scale=self.scale,
+            sparse_mode=3,
+        )
 
-            output, _ = torch_npu.npu_fused_infer_attention_score(
-                query=query,
-                key=key,
-                value=value,
-                atten_mask=attn_metadata.attn_mask,
-                block_table=attn_metadata.block_tables,
-                input_layout="TND",
-                block_size=block_size,
-                actual_seq_lengths=attn_metadata.actual_seq_lengths_q,
-                actual_seq_lengths_kv=attn_metadata.seq_lens_list,
-                num_key_value_heads=self.num_kv_heads,
-                num_heads=self.num_heads,
-                scale=self.scale,
-                sparse_mode=3,
-            )
-        else:
-            torch_npu._npu_paged_attention_splitfuse(
-                query=query,
-                key_cache=self.key_cache,
-                value_cache=self.value_cache,
-                mask=attn_metadata.attn_mask,
-                block_table=attn_metadata.block_tables,
-                seq_len=attn_metadata.query_lens,
-                context_lens=attn_metadata.seq_lens,
-                num_kv_heads=self.num_kv_heads,
-                num_heads=self.num_heads,
-                scale_value=self.scale,
-                out=output)
         return output
 
     def _attention_with_nomask_and_mask(self, q: torch.Tensor,
@@ -1155,12 +1142,11 @@ def forward(
                 query, attn_metadata, output)
         # Normal V1 situation.
         else:
-            if torch.version.cann.startswith("8.3"):
-                # npu_fused_infer_attention_score does not support cases
-                # where query.shape[0] != attn_metadata.query_start_loc[-1].
-                # Thus we need unpad it here.
-                num_tokens = attn_metadata.query_start_loc[-1]
-                query = query[:num_tokens]
+            # npu_fused_infer_attention_score does not support cases
+            # where query.shape[0] != attn_metadata.query_start_loc[-1].
+            # Thus we need unpad it here.
+            num_tokens = attn_metadata.query_start_loc[-1]
+            query = query[:num_tokens]
             intermediate_output = self._forward_v1_style(
                 query, attn_metadata, output)
 
diff --git a/vllm_ascend/ops/linear.py b/vllm_ascend/ops/linear.py
index cb738d10c12..eab312d5cf8 100644
--- a/vllm_ascend/ops/linear.py
+++ b/vllm_ascend/ops/linear.py
@@ -45,8 +45,8 @@ class AscendUnquantizedLinearMethod(UnquantizedLinearMethod):
 
     def process_weights_after_loading(self, layer: torch.nn.Module) -> None:
         super().process_weights_after_loading(layer)
-        if (is_enable_nz() and torch.version.cann.startswith("8.3") and
-                layer.weight.data.dtype in [torch.float16, torch.bfloat16]):
+        if (is_enable_nz() and layer.weight.data.dtype
+                in [torch.float16, torch.bfloat16]):
             layer.weight.data = torch_npu.npu_format_cast(
                 layer.weight.data, ACL_FORMAT_FRACTAL_NZ)
 
diff --git a/vllm_ascend/ops/linear_op.py b/vllm_ascend/ops/linear_op.py
index be7fa316bf2..1271f8e986f 100644
--- a/vllm_ascend/ops/linear_op.py
+++ b/vllm_ascend/ops/linear_op.py
@@ -411,9 +411,8 @@ def matmul_and_reduce(self, input_parallel: torch.Tensor,
                                                    quant_per_tensor)
 
         # For unquant
-        if mmrs_fusion and isinstance(
-                self.layer.quant_method, UnquantizedLinearMethod
-        ) and torch.version.cann.startswith("8.3"):
+        if mmrs_fusion and isinstance(self.layer.quant_method,
+                                      UnquantizedLinearMethod):
             output = torch_npu.npu_mm_reduce_scatter_base(
                 x,
                 self.layer.weight.t(),
@@ -429,8 +428,7 @@ def matmul_and_reduce(self, input_parallel: torch.Tensor,
         elif mmrs_fusion and (
                 isinstance(self.layer.quant_method, AscendLinearMethod)
                 and isinstance(self.layer.quant_method.quant_method,
-                               AscendW8A8LinearMethod)
-        ) and torch.version.cann.startswith("8.3"):
+                               AscendW8A8LinearMethod)):
             if x.dtype != torch.int8:
                 x_quant = quant_per_tensor(
                     x, self.layer.aclnn_input_scale_reciprocal,
diff --git a/vllm_ascend/worker/model_runner_v1.py b/vllm_ascend/worker/model_runner_v1.py
index a85d4cda026..a9ab3a4acde 100644
--- a/vllm_ascend/worker/model_runner_v1.py
+++ b/vllm_ascend/worker/model_runner_v1.py
@@ -367,13 +367,10 @@ def __init__(self, vllm_config: VllmConfig, device: torch.device):
                                              use_sparse=self.use_sparse)
         if self.pcp_size > 1:
             self.attn_mask_builder = None
-        elif torch.version.cann.startswith("8.3"):
+        else:
             self.attn_mask_builder = AttentionMaskBuilder(
                 self.scheduler_config.max_num_batched_tokens, self.dtype,
                 self.device)
-        else:
-            self.attn_mask_builder = AttentionMaskBuilder(
-                self.model_config.max_model_len, self.dtype)
 
         self._set_up_drafter()
 
@@ -988,11 +985,8 @@ def _make_attention_mask(self, seq_lens, position,
                 max_seq_len = max(seq_lens.max().item(), 0)
                 return self.attn_mask_builder.get_attn_mask(
                     max_seq_len, self.dtype, self.device)
-            elif torch.version.cann.startswith("8.3"):
-                return self.attn_mask_builder.get_splitfuse_attn_mask()
             else:
-                return self.attn_mask_builder.get_splitfuse_attn_mask(
-                    seq_lens, position, self.dtype, self.device)
+                return self.attn_mask_builder.get_splitfuse_attn_mask()
 
         # Prefill without cache situation.
         elif attn_state == AscendAttentionState.PrefillNoCache:
@@ -1001,12 +995,8 @@ def _make_attention_mask(self, seq_lens, position,
                 max_seq_len, self.dtype, self.device)
         # Prefill with cache hit.
         elif attn_state == AscendAttentionState.PrefillCacheHit:
-            if torch.version.cann.startswith("8.3"):
-                return self.attn_mask_builder.get_attn_mask(
-                    2048, self.dtype, self.device)
-            else:
-                return self.attn_mask_builder.get_attn_mask(
-                    128, self.dtype, self.device)
+            return self.attn_mask_builder.get_attn_mask(
+                2048, self.dtype, self.device)
         # Decode-only situation.
         else:
             return None