[Build] Force torch version (#3791)

We notice that sometimes user build vllm-ascend with incorrect torch
version. In this case, the build is passed, but when running the code,
the error `AttributeError: '_OpNamespace' '_C_ascend' object has no
attribute 'weak_ref_tensor'` is raised. Let's force the torch version to
2.7.1 and check the torch version when build from source to fix the
issue.

closes: #3342

- vLLM version: v0.11.0rc3
- vLLM main:
vllm-project/vllm@c9461e0

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>

Author: wangxiyuan

CMakeLists.txt

@@ -20,6 +20,13 @@ set(VLLM_ASCEND_INSTALL_PATH "${CMAKE_INSTALL_PREFIX}")
 
 find_package(Torch REQUIRED)
 
+run_python(TORCH_VERSION
+  "import torch; print(torch.__version__)" "Failed to locate torch path")
+# check torch version is 2.7.1
+if(NOT ${TORCH_VERSION} VERSION_EQUAL "2.7.1")
+  message(FATAL_ERROR "Expected PyTorch version 2.7.1, but found ${TORCH_VERSION}")
+endif()
+
 set(RUN_MODE "npu" CACHE STRING "cpu/sim/npu")
 set(SOC_VERSION ${SOC_VERSION})
 message(STATUS "Detected SOC version: ${SOC_VERSION}")

README.md

@@ -43,7 +43,7 @@ By using vLLM Ascend plugin, popular open-source models, including Transformer-l
 - 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))
-  * PyTorch >= 2.7.1, torch-npu >= 2.7.1.dev20250724
+  * PyTorch == 2.7.1, torch-npu == 2.7.1.dev20250724
   * vLLM (the same version as vllm-ascend)
 
 ## Getting Started

README.zh.md

@@ -44,7 +44,7 @@ vLLM 昇腾插件 (`vllm-ascend`) 是一个由社区维护的让vLLM在Ascend NP
 - 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))
-  * PyTorch >= 2.7.1, torch-npu >= 2.7.1.dev20250724
+  * PyTorch == 2.7.1, torch-npu == 2.7.1.dev20250724
   * vLLM (与vllm-ascend版本一致)
 
 ## 开始使用

docs/source/installation.md

@@ -13,8 +13,8 @@ This document describes how to install vllm-ascend manually.
     |---------------|----------------------------------|-------------------------------------------|
     | 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    |
-    | torch-npu     | >= 2.7.1.dev20250724             | 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           |
+    | torch-npu     | == 2.7.1.dev20250724             | 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           |
 
 There are two installation methods:
 - **Using pip**: first prepare env manually or via CANN image, then install `vllm-ascend` using pip.

examples/disaggregated_prefill_v1/mooncake_connector_deployment_guide.md

@@ -5,7 +5,7 @@
  *  Software:
      *  Python >= 3.9, < 3.12
      *  CANN >= 8.2.rc1
-     *  PyTorch >= 2.7.1, torch-npu >= 2.7.1.dev20250724
+     *  PyTorch == 2.7.1, torch-npu == 2.7.1.dev20250724
      *  vLLM (same version as vllm-ascend)
      *  mooncake-transfer-engine reference documentation: https://github.com/kvcache-ai/Mooncake/blob/main/doc/zh/ascend_transport.md
 

examples/disaggregated_prefill_v1/mooncake_connector_store_deployment_guide.md

@@ -5,7 +5,7 @@
 * Software:
   * Python >= 3.9, < 3.12
   * CANN >= 8.2.rc1
-  * PyTorch >= 2.7.1, torch-npu >= 2.7.1.dev20250724
+  * PyTorch == 2.7.1, torch-npu == 2.7.1.dev20250724
   * vLLM：main branch
   * vLLM-Ascend：main branch
   * Mooncake：[AscendTransport/Mooncake at pooling-async-memcpy](https://github.com/AscendTransport/Mooncake/tree/pooling-async-memcpy)(Currently available branch code, continuously updated.)

pyproject.toml

@@ -15,7 +15,7 @@ requires = [
     "setuptools>=64",
     "setuptools-scm>=8",
     "torch-npu==2.7.1.dev20250724",
-    "torch>=2.7.1",
+    "torch==2.7.1",
     "torchvision",
     "wheel",
     "msgpack",

requirements.txt

@@ -11,7 +11,7 @@ scipy
 pandas
 setuptools>=64
 setuptools-scm>=8
-torch>=2.7.1
+torch==2.7.1
 torchvision
 wheel
 pandas-stubs