From c3d8fcabc361faa6697c87d64d088349da7b4127 Mon Sep 17 00:00:00 2001 From: zihao-intuition Date: Tue, 17 Jun 2025 16:29:20 +0800 Subject: [PATCH] faq Update docs/msadapter/docs/source_zh_cn/note/pytorch_api_supporting_overview.md fix --- .../note/pytorch_api_supporting_overview.md | 15 +++++++++++++++ 1 file changed, 15 insertions(+) diff --git a/docs/msadapter/docs/source_zh_cn/note/pytorch_api_supporting_overview.md b/docs/msadapter/docs/source_zh_cn/note/pytorch_api_supporting_overview.md index 5d9c3a842e..171ab6b448 100644 --- a/docs/msadapter/docs/source_zh_cn/note/pytorch_api_supporting_overview.md +++ b/docs/msadapter/docs/source_zh_cn/note/pytorch_api_supporting_overview.md @@ -232,6 +232,21 @@ after import torch: a.shape=torch.Size([2]) |------|-----------| | mindspore.Tensor/mindspore.StubTensor | is_shared, softmax, type_, retain_grad, shape, to_dense, \_base, data, numel, nelement, repeat, cuda, npu, cpu, size, dim, clone, log_softmax, narrow, view, \_\_or\_\_, device, \_\_and\_\_, \_\_xor\_\_, \_\_iter\_\_, \_\_reduce_ex\_\_, expand, detach, T, transpose, mean, clamp, is_cuda, is_cpu, repeat_interleave, is_sparse, requires_grad, requires_grad_, unsqueeze, \_\_pow\_\_, float, backward, split, norm, record_stream, data_ptr, pin_memory, grad, \_\_imul\_\_, reshape, squeeze, element_size, exponential\_ | +### 不支持非PyTorch API官方使用方式 + +PyTorch通过pybind重载的场景下，当*号前存在key=value的默认值时，参数可以显式通过size=[1,2]的形式传参。 + +MSAdapter目前不支持此类传参方式，并且这类传参方式与PyTorch官方使用不符。 + +以torch.randint为例 + +```python +label = torch.randint(0, 10, size=[1, 2], dtype=torch.float) # PyTorch可运行，但是MSAdapter报错；非PyTorch官网使用方式 +label = torch.randint(0, 10, [1, 2], dtype=torch.float) # PyTorch，MSAdapter均可运行；遵照PyTorch官网使用方式 +``` + +用户需要按照PyTorch官方的方法使用参数，MSAdapter已和PyTorch官网的用法对齐。 + ## PyTorch模块支持情况 用户可以参考下列表格查看各模块API支持情况。 -- Gitee