有没有傻瓜式的一键安装的方法？

[ZhaoVenya]

🚀 The feature, motivation and pitch

我用的版本是：triton_v3.5.x
我是按照这个文档的从源代码安装这个步骤来的。
执行过程如下：
export FLAGTREE_BACKEND=nvidia # 当是nvidia的时候这个可以不用设置? cd python # Need to enter the python directory for Triton 3.1/3.2/3.3 python3 -m pip install . --no-build-isolation -v # Install flagtree and uninstall triton python3 -m pip show flagtree cd ${ANY_OTHER_PATH}; python3 -c 'import triton; print(triton.__path__)'

但是一直报错 根本解决不完。。。

源代码安装之前，需要下载什么开源工程吗？比如llvm-project、mlir、triton这些是否需要提前下载好？对应的版本是否有特殊要求？
小白确实不知道怎么下手T_T，跪求大佬们给小白一个傻瓜式的安装教程

Alternatives

No response

Additional context

No response

[sgjzfzzf]

Hi @ZhaoVenya! Thank you for your attention to FlagOS. For installation, it's almost the same as installing it on Nvidia as triton. My experience is to run pip install -v . in the root directory on the triton_v3.5.x branch. It would be easier to start from a new environment and check whether the installation succeeds by running python -c "import triton". Hope this could help you! For other platforms, please refer to the relevant docs.

If you still have problems, you could paste them and the fatal reports in the issue. Please know that English is preferred for our global community.

[zhzhcookie]

很抱歉给您带来了不好的安装体验，我们将会持续更新文档说明，改善易用性。

以 ubuntu 系统环境下的英伟达显卡为例，如果您具备良好的国际网环境，那么首页上展示的命令即可完成构建安装：

apt install zlib1g zlib1g-dev libxml2 libxml2-dev
python3 -m pip install -r python/requirements.txt
cd python
python3 -m pip install . --no-build-isolation -v

如果不具备，那么可以参考 https://github.com/flagos-ai/FlagTree/blob/main/documents/build_cn.md#-nvidia--amd-nvidia--amd
有四个步骤可以顺序执行，从下载镜像、安装依赖到执行构建安装命令。

如果希望免源码安装，可以在下载镜像之后，参考 https://github.com/flagos-ai/FlagTree/blob/main/README_cn.md#%E5%85%8D%E6%BA%90%E7%A0%81%E5%AE%89%E8%A3%85
执行命令：

python3 -m pip uninstall -y triton  # Repeat the cmd until fully uninstalled
RES="--index-url=https://resource.flagos.net/repository/flagos-pypi-hosted/simple"
python3 -m pip install flagtree==0.5.0 $RES

[gbdjxgp]

@zhzhcookie 您好，我这边是昇腾环境，如何校验安装的是否正确？readme里面指向的unittest（cd third_party/${backend_name}/python/test/unit）目录是错的，我切换到图中的路径还是运行报错，我是用pypi安装的flagtree==0.5.0+ascend3.2

[zhzhcookie]

@zhzhcookie 您好，我这边是昇腾环境，如何校验安装的是否正确？readme里面指向的unittest（cd third_party/${backend_name}/python/test/unit）目录是错的，我切换到图中的路径还是运行报错，我是用pypi安装的flagtree==0.5.0+ascend3.2

实际的后端测试用例可参考 https://github.com/flagos-ai/FlagTree/blob/triton_v3.2.x/.github/workflows/ascend-build-and-test.yml

source /usr/local/Ascend/ascend-toolkit/set_env.sh

pushd third_party/ascend/tutorials
python3 01-vector-add.py
python3 02-fused-softmax.py
popd

pushd third_party/ascend/unittest/pytest_ut
python3 -m pytest . \
    --ignore=test_add_mindspore.py \
    --ignore=test_softmax_mindspore.py \
    --ignore=test_assume.py \
    --ignore=test_index_select.py
popd

[gbdjxgp]

@zhzhcookie
感谢回复！
您好，我按照您的方式操作，前两个py文件运行成功了，但unittest运行全是图中的错误，看了下应该是缺少依赖：pytest-xdist装完后我这边也能成功跑起来啦，但运行到最后还是有一些错误：

python -m pytest -s test_assume.py test_hoistbroadcast.py test_index_select.py test_nextafter.py

E               triton.compiler.errors.MLIRCompilationError: 
E               ///------------------[ERROR][Triton][BEG]------------------
E               [ConvertLinalgRToBinary] encounters error:
E               [Errno 2] No such file or directory: '/tmp/tmp216ucbf1/kernel.o'
E               
E               [INFO]: The compiled kernel cache is in /root/.triton/cache/kmU_D88ek2KpgMTiYz8MruKiy2REPu-bDR0IgrExW98
E               
E               ///------------------[ERROR][Triton][END]------------------

../../../../.venv/lib/python3.11/site-packages/triton/backends/ascend/spec/triton/compiler/compiler.py:295: MLIRCompilationError
=========================================================================== short test summary info ============================================================================
FAILED test_assume.py::test_assume[param_list0] - triton.compiler.errors.MLIRCompilationError: 
FAILED test_hoistbroadcast.py::test_copy[param_list0] - AssertionError: Tensor-likes are not close!
FAILED test_hoistbroadcast.py::test_hoistbroadcast_compare[param_list0] - AssertionError: Tensor-likes are not close!
FAILED test_index_select.py::test_index_select_extension[src_shape0-0-indice_shape0-float32] - triton.compiler.errors.MLIRCompilationError: 
FAILED test_index_select.py::test_index_select_extension[src_shape1-0-indice_shape1-float32] - triton.compiler.errors.MLIRCompilationError: 
FAILED test_index_select.py::test_index_select_extension[src_shape2-0-indice_shape2-float32] - triton.compiler.errors.MLIRCompilationError: 
FAILED test_index_select.py::test_index_select_extension[src_shape3-0-indice_shape3-float32] - triton.compiler.errors.MLIRCompilationError: 
FAILED test_index_select.py::test_index_select_extension[src_shape4-0-indice_shape4-float32] - triton.compiler.errors.MLIRCompilationError: 
FAILED test_index_select.py::test_index_select_extension[src_shape5-0-indice_shape5-float32] - triton.compiler.errors.MLIRCompilationError: 
FAILED test_index_select.py::test_index_select_extension[src_shape6-0-indice_shape6-float32] - triton.compiler.errors.MLIRCompilationError: 
FAILED test_index_select.py::test_index_select_extension[src_shape7-0-indice_shape7-float32] - triton.compiler.errors.MLIRCompilationError: 
FAILED test_index_select.py::test_index_select_extension[src_shape8-0-indice_shape8-float32] - triton.compiler.errors.MLIRCompilationError: 
FAILED test_index_select.py::test_index_select_extension[src_shape9-0-indice_shape9-float32] - triton.compiler.errors.MLIRCompilationError: 
FAILED test_index_select.py::test_index_select_extension[src_shape10-0-indice_shape10-float32] - triton.compiler.errors.MLIRCompilationError: 
FAILED test_index_select.py::test_index_select_extension[src_shape11-0-indice_shape11-float32] - triton.compiler.errors.MLIRCompilationError: 
FAILED test_index_select.py::test_index_select_extension[src_shape12-0-indice_shape12-float32] - triton.compiler.errors.MLIRCompilationError: 
FAILED test_index_select.py::test_index_select_extension[src_shape13-0-indice_shape13-float32] - triton.compiler.errors.MLIRCompilationError: 
FAILED test_index_select.py::test_index_select_extension[src_shape14-0-indice_shape14-float32] - triton.compiler.errors.MLIRCompilationError: 
FAILED test_index_select.py::test_index_select_extension[src_shape15-0-indice_shape15-float32] - triton.compiler.errors.MLIRCompilationError: 
FAILED test_index_select.py::test_index_select_extension[src_shape16-0-indice_shape16-float32] - triton.compiler.errors.MLIRCompilationError: 
FAILED test_index_select.py::test_index_select_extension[src_shape17-0-indice_shape17-float32] - triton.compiler.errors.MLIRCompilationError: 
FAILED test_nextafter.py::test_nextafter[param_list2] - triton.compiler.errors.MLIRCompilationError: 
=================================================================== 22 failed, 39 passed in 99.48s (0:01:39) ===================================================================

请问日志中的这种错误会影响使用吗?
还有一个问题：这个项目的定位就是可以在各个国产硬件上运行triton/triton TLE对么，在昇腾上是不是可以理解为triton-ascend的一种替代方案？

[zhzhcookie]

@zhzhcookie 感谢回复！ 您好，我按照您的方式操作，前两个py文件运行成功了，但unittest运行全是图中的错误，看了下应该是缺少依赖：pytest-xdist装完后我这边也能成功跑起来啦，但运行到最后还是有一些错误：

python -m pytest -s test_assume.py test_hoistbroadcast.py test_index_select.py test_nextafter.py

E               triton.compiler.errors.MLIRCompilationError: 
E               ///------------------[ERROR][Triton][BEG]------------------
E               [ConvertLinalgRToBinary] encounters error:
E               [Errno 2] No such file or directory: '/tmp/tmp216ucbf1/kernel.o'
E               
E               [INFO]: The compiled kernel cache is in /root/.triton/cache/kmU_D88ek2KpgMTiYz8MruKiy2REPu-bDR0IgrExW98
E               
E               ///------------------[ERROR][Triton][END]------------------

../../../../.venv/lib/python3.11/site-packages/triton/backends/ascend/spec/triton/compiler/compiler.py:295: MLIRCompilationError
=========================================================================== short test summary info ============================================================================
FAILED test_assume.py::test_assume[param_list0] - triton.compiler.errors.MLIRCompilationError: 
FAILED test_hoistbroadcast.py::test_copy[param_list0] - AssertionError: Tensor-likes are not close!
FAILED test_hoistbroadcast.py::test_hoistbroadcast_compare[param_list0] - AssertionError: Tensor-likes are not close!
FAILED test_index_select.py::test_index_select_extension[src_shape0-0-indice_shape0-float32] - triton.compiler.errors.MLIRCompilationError: 
FAILED test_index_select.py::test_index_select_extension[src_shape1-0-indice_shape1-float32] - triton.compiler.errors.MLIRCompilationError: 
FAILED test_index_select.py::test_index_select_extension[src_shape2-0-indice_shape2-float32] - triton.compiler.errors.MLIRCompilationError: 
FAILED test_index_select.py::test_index_select_extension[src_shape3-0-indice_shape3-float32] - triton.compiler.errors.MLIRCompilationError: 
FAILED test_index_select.py::test_index_select_extension[src_shape4-0-indice_shape4-float32] - triton.compiler.errors.MLIRCompilationError: 
FAILED test_index_select.py::test_index_select_extension[src_shape5-0-indice_shape5-float32] - triton.compiler.errors.MLIRCompilationError: 
FAILED test_index_select.py::test_index_select_extension[src_shape6-0-indice_shape6-float32] - triton.compiler.errors.MLIRCompilationError: 
FAILED test_index_select.py::test_index_select_extension[src_shape7-0-indice_shape7-float32] - triton.compiler.errors.MLIRCompilationError: 
FAILED test_index_select.py::test_index_select_extension[src_shape8-0-indice_shape8-float32] - triton.compiler.errors.MLIRCompilationError: 
FAILED test_index_select.py::test_index_select_extension[src_shape9-0-indice_shape9-float32] - triton.compiler.errors.MLIRCompilationError: 
FAILED test_index_select.py::test_index_select_extension[src_shape10-0-indice_shape10-float32] - triton.compiler.errors.MLIRCompilationError: 
FAILED test_index_select.py::test_index_select_extension[src_shape11-0-indice_shape11-float32] - triton.compiler.errors.MLIRCompilationError: 
FAILED test_index_select.py::test_index_select_extension[src_shape12-0-indice_shape12-float32] - triton.compiler.errors.MLIRCompilationError: 
FAILED test_index_select.py::test_index_select_extension[src_shape13-0-indice_shape13-float32] - triton.compiler.errors.MLIRCompilationError: 
FAILED test_index_select.py::test_index_select_extension[src_shape14-0-indice_shape14-float32] - triton.compiler.errors.MLIRCompilationError: 
FAILED test_index_select.py::test_index_select_extension[src_shape15-0-indice_shape15-float32] - triton.compiler.errors.MLIRCompilationError: 
FAILED test_index_select.py::test_index_select_extension[src_shape16-0-indice_shape16-float32] - triton.compiler.errors.MLIRCompilationError: 
FAILED test_index_select.py::test_index_select_extension[src_shape17-0-indice_shape17-float32] - triton.compiler.errors.MLIRCompilationError: 
FAILED test_nextafter.py::test_nextafter[param_list2] - triton.compiler.errors.MLIRCompilationError: 
=================================================================== 22 failed, 39 passed in 99.48s (0:01:39) ===================================================================

请问日志中的这种错误会影响使用吗? 还有一个问题：这个项目的定位就是可以在各个国产硬件上运行triton/triton TLE对么，在昇腾上是不是可以理解为triton-ascend的一种替代方案？

感谢关注。
测例中一些测试没跑过，Tensor-likes are not close 含义是输出值精度不达标，MLIRCompilationError: 含义是编译降级失败，这表明完备度建设还需持续进行，也欢迎社区用户提供修复意见。
FlagTree 的 ascend 后端与 triton-ascend 是合作共建关系，华为是 flagos 社区的贡献成员之一，后者更关注昇腾芯片上的开法使用，前者更关注多芯片整体生态、组件和优化手段复用，使得不同后端使用同一套编程语言接口。