From fe5dfeee757c0ee1f90fc70ec7dfed2b06e55224 Mon Sep 17 00:00:00 2001 From: huangbingjian Date: Fri, 14 May 2021 15:10:22 +0800 Subject: [PATCH] add backend_running faq --- docs/faq/source_en/backend_running.md | 8 ++++++++ docs/faq/source_zh_cn/backend_running.md | 6 ++++++ 2 files changed, 14 insertions(+) diff --git a/docs/faq/source_en/backend_running.md b/docs/faq/source_en/backend_running.md index 54d4abc413..58a491022f 100644 --- a/docs/faq/source_en/backend_running.md +++ b/docs/faq/source_en/backend_running.md @@ -222,3 +222,11 @@ A: This message means that MindSpore failed to load library `libgpu_collective.s A：Normally, GPU version of MindSpore doesn't need to set `DEVICE_ID`. MindSpore automatically chooses visible GPU devices according to the cuda environment variable `CUDA_VISIBLE_DEVICES`. After setting `CUDA_VISIBLE_DEVICES`, `DEVICE_ID` refers to the ordinal of the GPU device: - After `export CUDA_VISIBLE_DEVICES=1,3,5`, `DEVICE_ID` should be exported as `0`, `1` or `2`. If `3` is exported, MindSpore will fail to execute because of the invalid device ordinal. + +
+ +**Q：After running the script, an error of `Type join failed` appears, how to solve it？** + +A：In the static analysis step, the abstracts of multiple nodes will be joined. If their types are inconsistent, an error similar to `Type join failed, type1 = Int64, type2 = Int32` will occur. This problem may be caused by incorrect data type settings in the script, and please pay attention to the default data type. You can locate the problem through the ERROR information such as node, evaluator, abstract, etc., and find nodes with inconsistent types through the `analyze_fail.dat` file. + +
diff --git a/docs/faq/source_zh_cn/backend_running.md b/docs/faq/source_zh_cn/backend_running.md index f547f331a4..8273829d3d 100644 --- a/docs/faq/source_zh_cn/backend_running.md +++ b/docs/faq/source_zh_cn/backend_running.md @@ -292,3 +292,9 @@ A：MindSpore GPU模式一般无需设置`DEVICE_ID`环境变量，MindSpore会 - 执行`export CUDA_VISIBLE_DEVICES=1,3,5`后，`DEVICE_ID`应当被设置为`0`，`1`或`2`，若设置为`3`及以上，MindSpore会由于设备ID不合法而运行失败。
+ +**Q：运行脚本出现`Type join failed`的报错，如何解决？** + +A：在静态分析步骤中，对多个节点的abstract进行合并，如果abstract的type类型不一致，则会出现类似于`Type join failed, type1 = Int64, type2 = Int32`的报错。该问题可能由脚本中的数据类型设置不正确引起，请留意默认数据类型。可以通过紧接着的node、evaluator、abstract等ERROR报错信息定位问题，并从`analyze_fail.dat`文件中找到type类型不一致的节点。 + +
-- Gitee