新增MultiLabelMarginLoss #73538
新增MultiLabelMarginLoss #73538
Conversation
modified: python/paddle/nn/__init__.py modified: python/paddle/nn/functional/__init__.py modified: python/paddle/nn/functional/loss.py modified: python/paddle/nn/layer/__init__.py modified: python/paddle/nn/layer/loss.py new file: test/legacy_test/test_multilabelmarginloss.py
|
你的PR提交成功，感谢你对开源项目的贡献! |
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## develop #73538 +/- ##
==========================================
Coverage ? 95.74%
==========================================
Files ? 2
Lines ? 47
Branches ? 0
==========================================
Hits ? 45
Misses ? 2
Partials ? 0 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
modified: test/legacy_test/CMakeLists.txt modified: test/legacy_test/test_multilabelmarginloss.py
|
docs: PaddlePaddle/docs#7331 |
python/paddle/nn/functional/loss.py
Outdated
| sample_loss = paddle.zeros([], dtype=input.dtype) | ||
|
|
||
| # check each position for valid label | ||
| for j in range(dim): |
There was a problem hiding this comment.
这里的公式看起来比multi_label_soft_margin_loss要复杂很多，使用的n*n复杂度计算，对性能的影响也比较大，单测超时了。
这里与multi_label_soft_margin_loss的公式差异很大吗，能否有更简洁高效的实现
There was a problem hiding this comment.
嗯，收到，因为静态图无法用if else，所以修改了很多，那我改成底层的cpp和cuda实现吧
|
Sorry to inform you that abd1da5's CIs have passed for more than 7 days. To prevent PR conflicts, you need to re-run all CIs manually. |
modified: python/paddle/nn/functional/loss.py
|
已修改一版新的函数 |
|
CI报错： 2025-07-04 08:19:58 /paddle/paddle/phi/kernels/funcs/fft_fill_conj.h(163): error: identifier "ctx" is undefined
2025-07-04 08:19:58 detected during instantiation of "void phi::FFTR2CKernel<T,Context>(const Context &, const phi::DenseTensor &, const std::vector<int64_t, std::allocator<int64_t>> &, const std::__cxx11::string &, __nv_bool, __nv_bool, phi::DenseTensor *) [with T=float, Context=phi::GPUContext]"
2025-07-04 08:19:58 /paddle/paddle/phi/kernels/gpu/fft_kernel.cu(34): here
2025-07-04 08:19:58
2025-07-04 08:19:58 /paddle/paddle/phi/kernels/funcs/fft_fill_conj.h(163): error: type name is not allowed
2025-07-04 08:19:58 detected during instantiation of "void phi::FFTR2CKernel<T,Context>(const Context &, const phi::DenseTensor &, const std::vector<int64_t, std::allocator<int64_t>> &, const std::__cxx11::string &, __nv_bool, __nv_bool, phi::DenseTensor *) [with T=float, Context=phi::GPUContext]"
2025-07-04 08:19:58 /paddle/paddle/phi/kernels/gpu/fft_kernel.cu(34): here应该和pr修改部分无关 猜测可能和这个pr有关 |
|
PaConvert: PaddlePaddle/PaConvert#599 |
test/legacy_test/CMakeLists.txt
Outdated
| @@ -1247,3 +1247,4 @@ set_tests_properties(test_imperative_selected_rows_to_lod_tensor | |||
| PROPERTIES TIMEOUT 200) | |||
| set_tests_properties(test_slice_op PROPERTIES TIMEOUT 120) | |||
| set_tests_properties(test_load_state_dict_from_url PROPERTIES TIMEOUT 40) | |||
| set_tests_properties(test_multimarginloss PROPERTIES TIMEOUT 120) | |||
|
本地测试时间 Ran 4 tests in 0.192s |
PR Category
User Experience
PR Types
Improvements
Description
当前问题
目前paddle没有MultiLabelMarginLoss
解决方案
新增MultiLabelMarginLoss