GitHub / pytorch/pytorch issues and pull requests

Labelled with: module: nn

#165154 - Add scaled_grouped_mm_v2 and python API

Pull Request - State: open - Opened by slayton58 about 1 month ago - 8 comments
Labels: module: nn, ciflow/trunk, release notes: quantization, topic: new features, merging, ciflow/rocm, ciflow/h100, ciflow/b200

#163918 - DISABLED test_TransformerEncoderLayer_relu_activation_cuda_tf32 (main.TestNN)

Issue - State: open - Opened by pytorch-bot[bot] about 2 months ago
Labels: module: nn, module: rocm, triaged, module: flaky-tests, skipped

#163917 - DISABLED test_TransformerDecoderLayer_relu_activation_cuda_tf32 (main.TestNN)

Issue - State: open - Opened by pytorch-bot[bot] about 2 months ago
Labels: module: nn, module: rocm, triaged, module: flaky-tests, skipped

#163640 - [torch.compile] TransformerEncoder with boolean src_key_padding_mask fails under Dynamo fullgraph mode

Issue - State: closed - Opened by tinywisdom about 2 months ago - 2 comments
Labels: module: nn, module: error checking, triaged, oncall: pt2

#162363 - [Conv1d] Check overflow before we compute padding size.

Pull Request - State: open - Opened by thenumberouscode 2 months ago - 21 comments
Labels: module: nn, module: cpu, triaged, open source, release notes: nn

#161252 - Improve doc examples for softmax

Issue - State: closed - Opened by owhelloworld 3 months ago - 2 comments
Labels: module: docs, module: nn, triaged

#160952 - [Bilinear] move check to reset_parameters

Pull Request - State: closed - Opened by parsshar-RH 3 months ago - 8 comments
Labels: module: nn, triaged, open source, Merged, ciflow/trunk, topic: not user facing, merging

#160407 - `Bilinear` module can't be made lazy due to new checks introduced in `init`

Issue - State: closed - Opened by Rathur421 3 months ago - 6 comments
Labels: module: nn, triaged

#160053 - `torch.Pad(mode="circular")` doesn't work for 4D or 5D input despite error msg

Issue - State: closed - Opened by JulianKnodt 3 months ago - 2 comments
Labels: module: nn, triaged, actionable

#158345 - @cached_property doesn't raise exception in torch.nn.Module

Issue - State: closed - Opened by Tytskiy 4 months ago - 3 comments
Labels: module: nn, triaged

#157663 - ConvNd ops in channel last layout (N,L,C) / (N,H,W,C) / (N,D,H,W,C)

Issue - State: closed - Opened by mauriceweiler 5 months ago - 1 comment
Labels: module: nn, triaged, module: memory format

#157259 - DISABLED test_affine_3d_rotateRandom_cuda (main.TestNNDeviceTypeCUDA)

Issue - State: closed - Opened by pytorch-bot[bot] 5 months ago - 2 comments
Labels: module: nn, module: rocm, triaged, module: flaky-tests, skipped

#157167 - DISABLED test_TransformerEncoderLayer_gelu_activation_cuda_fp32 (main.TestNN)

Issue - State: open - Opened by pytorch-bot[bot] 5 months ago
Labels: module: nn, module: rocm, triaged, module: flaky-tests, skipped

#157166 - DISABLED test_linear_autograd_device_cuda_nobias_weightCSR (main.TestNN)

Issue - State: open - Opened by pytorch-bot[bot] 5 months ago
Labels: module: nn, module: rocm, triaged, module: flaky-tests, skipped

#157165 - DISABLED test_linear_autograd_device_cuda_bias_weightCOO (main.TestNN)

Issue - State: open - Opened by pytorch-bot[bot] 5 months ago
Labels: module: nn, module: rocm, triaged, module: flaky-tests, skipped

#157164 - DISABLED test_grid_sample_large_cuda (main.TestNNDeviceTypeCUDA)

Issue - State: open - Opened by pytorch-bot[bot] 5 months ago
Labels: module: nn, module: rocm, triaged, module: flaky-tests, skipped

#157121 - DISABLED test_TransformerDecoderLayer_gelu_activation_cuda_fp32 (main.TestNN)

Issue - State: open - Opened by pytorch-bot[bot] 5 months ago
Labels: module: nn, module: rocm, triaged, module: flaky-tests, skipped

#157118 - DISABLED test_Linear_cuda_fp32 (main.TestNN)

Issue - State: open - Opened by pytorch-bot[bot] 5 months ago
Labels: module: nn, module: rocm, triaged, module: flaky-tests, skipped

#157115 - DISABLED test_Conv2d_depthwise_dilated_cuda_fp32 (main.TestNN)

Issue - State: open - Opened by pytorch-bot[bot] 5 months ago
Labels: module: nn, module: rocm, triaged, module: flaky-tests, skipped

#157113 - DISABLED test_affine_2d_rotate45_cuda (main.TestNNDeviceTypeCUDA)

Issue - State: closed - Opened by pytorch-bot[bot] 5 months ago - 4 comments
Labels: module: nn, module: rocm, triaged, module: flaky-tests, skipped

#157083 - DISABLED test_TransformerEncoderLayer_relu_activation_cuda_fp32 (main.TestNN)

Issue - State: open - Opened by pytorch-bot[bot] 5 months ago
Labels: module: nn, module: rocm, triaged, module: flaky-tests, skipped

#157082 - DISABLED test_rnn_fused_cuda_float32 (main.TestNNDeviceTypeCUDA)

Issue - State: open - Opened by pytorch-bot[bot] 5 months ago
Labels: module: nn, module: rocm, triaged, module: flaky-tests, skipped

#157081 - DISABLED test_Linear_no_batch_dim_cuda_fp32 (main.TestNN)

Issue - State: closed - Opened by pytorch-bot[bot] 5 months ago - 3 comments
Labels: module: nn, module: rocm, triaged, module: flaky-tests, skipped

#157080 - DISABLED test_linear_autograd_device_cuda_nobias_weightCSC (main.TestNN)

Issue - State: closed - Opened by pytorch-bot[bot] 5 months ago - 2 comments
Labels: module: nn, module: rocm, triaged, module: flaky-tests, skipped

#157079 - DISABLED test_linear_autograd_device_cuda_bias_weightCSC (main.TestNN)

Issue - State: open - Opened by pytorch-bot[bot] 5 months ago
Labels: module: nn, module: rocm, triaged, module: flaky-tests, skipped

#157001 - DISABLED test_Conv2d_size_1_kernel_cuda (main.TestConvolutionNNDeviceTypeCUDA)

Issue - State: open - Opened by pytorch-bot[bot] 5 months ago
Labels: module: nn, module: rocm, triaged, module: flaky-tests, skipped

#157000 - DISABLED MULTIPLE There are multiple flaky tests in test_nn.py

Issue - State: closed - Opened by pytorch-bot[bot] 5 months ago - 1 comment
Labels: module: nn, module: rocm, triaged, module: flaky-tests, skipped, aggregate flaky test issue

#156740 - Add explicit typing to nn.Module init()

Issue - State: closed - Opened by gabriele-marino 5 months ago - 3 comments
Labels: module: nn, module: typing, triaged

#156695 - Hook can only collect positional arguments, but not hook keyword arguments

Issue - State: closed - Opened by caoyang-sufe 5 months ago - 2 comments
Labels: module: nn, triaged

#155225 - canUse32BitIndexMath set to False with efficient net

Issue - State: closed - Opened by jjh42 6 months ago - 13 comments
Labels: module: nn, module: cuda, triaged

#155217 - DISABLED test_TransformerEncoderLayer_gelu_activation_cuda_tf32 (main.TestNN)

Issue - State: open - Opened by pytorch-bot[bot] 6 months ago - 3 comments
Labels: module: nn, module: rocm, triaged, module: flaky-tests, skipped

#155216 - DISABLED test_Linear_cuda_tf32 (main.TestNN)

Issue - State: closed - Opened by pytorch-bot[bot] 6 months ago - 6 comments
Labels: high priority, triage review, module: nn, module: rocm, triaged, module: flaky-tests, skipped

#153620 - Update rnn.py, fix `torch.nn.RNN` document error

Pull Request - State: closed - Opened by AIboy996 6 months ago - 12 comments
Labels: module: docs, module: nn, open source, ciflow/trunk, topic: not user facing, merging

#153591 - `torch.nn.functional.interpolate` doesn't have `[source]`

Issue - State: closed - Opened by hyperkai 6 months ago - 1 comment
Labels: module: docs, module: nn, triaged

#153281 - `torch.nn.functional.rrelu` crashes on CPU with `training=True` when `lower` or `upper` is set to `inf`

Issue - State: closed - Opened by jiren-the-gray 6 months ago - 3 comments
Labels: module: nn, triaged, module: NaNs and Infs, module: edge cases

#153137 - `torch.batch_norm` shows inconsistent error behavior between CPU and GPU

Issue - State: closed - Opened by SilentTester73 7 months ago - 1 comment
Labels: module: nn, module: error checking, triaged, module: norms and normalization, module: edge cases

#152156 - Note some limit in docstring of `padding` in Poolnd

Issue - State: closed - Opened by ILCSFNO 7 months ago - 1 comment
Labels: module: docs, module: nn, triaged

#151959 - [FlexAttention] Remove Old Constraint on lastdim strides

Pull Request - State: closed - Opened by drisspg 7 months ago - 11 comments
Labels: module: nn, Merged, ciflow/trunk, release notes: nn, topic: bug fixes, module: inductor, ciflow/inductor, merging

#151103 - Whether `recompute_scale_factor=True` needs `scale_factor` passed in or not in `torch.nn.Upsample()`?

Issue - State: closed - Opened by ILCSFNO 7 months ago - 2 comments
Labels: module: nn, triaged

#150995 - `torch.nn.functional.ctc_loss` inconsistent implementation and docs

Issue - State: closed - Opened by zeshengzong 7 months ago
Labels: module: nn, triaged

#150835 - `Floating point exception` in `torch.nn.functional.ctc_loss`

Issue - State: closed - Opened by vwrewsge 8 months ago
Labels: module: nn, module: cuda, triaged, module: edge cases

#150819 - Optimize `ConvTranspose2d` stride description

Pull Request - State: closed - Opened by zeshengzong 8 months ago - 10 comments
Labels: module: nn, module: convolution, triaged, open source, Merged, ciflow/trunk, release notes: nn, topic: docs, merging

#150775 - ConvTranspose2d documentation should clarify behavior of stride > 1 (zero insertion)

Issue - State: closed - Opened by EduardoLawson1 8 months ago - 2 comments
Labels: module: docs, module: nn, module: convolution, triaged, actionable

#150729 - [BE] Add `all` to `torch/nn/functional.pyi` and `torch/return_types.pyi`

Pull Request - State: open - Opened by XuehaiPan 8 months ago - 4 comments
Labels: module: nn, open source, better-engineering, module: codegen, topic: not user facing

#150729 - [BE] Add `all` to `torch/nn/functional.pyi` and `torch/return_types.pyi`

Pull Request - State: closed - Opened by XuehaiPan 8 months ago - 9 comments
Labels: module: nn, open source, better-engineering, module: codegen, Merged, ciflow/trunk, topic: not user facing, merging

#150404 - Running `LazyModuleMixin` example throw errors

Issue - State: closed - Opened by zeshengzong 8 months ago
Labels: module: nn, triaged

#150170 - `torch.nn.Threshold` and `torch.nn.GLU` missing plot

Issue - State: closed - Opened by zeshengzong 8 months ago
Labels: module: docs, module: nn, triaged

#150016 - [ROCm] RuntimeError: HIPBLAS_STATUS_NOT_SUPPORTED in torch 2.6.0+rocm6.2.4

Issue - State: closed - Opened by zhaohm14 8 months ago - 13 comments
Labels: high priority, module: nn, module: rocm, triaged

#149886 - TransformerDecoder produces identical outputs regardless of input

Issue - State: closed - Opened by nicolacalzone 8 months ago
Labels: module: nn, triaged

#149853 - [nn] Implement PartialLinear module for structured sparsity

Pull Request - State: closed - Opened by lakshminarasimmanv 8 months ago - 14 comments
Labels: feature, module: nn, triaged, open source, topic: not user facing

#149693 - Fix broken LazyLinear init

Pull Request - State: closed - Opened by vmoens 8 months ago - 7 comments
Labels: module: nn, Merged, ciflow/trunk, release notes: nn, merging

#149691 - LazyLinear broken by new init logic

Issue - State: closed - Opened by vmoens 8 months ago - 2 comments
Labels: high priority, module: nn, triaged, module: regression, module: lazy

#149604 - DISABLED test_linear (main.TestLazyModules)

Issue - State: closed - Opened by pytorch-bot[bot] 8 months ago - 2 comments
Labels: module: nn, triaged, module: flaky-tests, skipped

#149534 - Addind RoPE to pytorch core

Issue - State: open - Opened by manuelcandales 8 months ago
Labels: module: nn

#149474 - nn.Linear layer initialization formula wrong in docs

Issue - State: closed - Opened by karanjakhar 8 months ago - 2 comments
Labels: module: docs, module: nn, triaged, actionable

#149134 - add keepdim to cosine similarity

Pull Request - State: open - Opened by Isalia20 8 months ago - 4 comments
Labels: module: nn, triaged, open source, release notes: onnx, topic: improvements

#149134 - add keepdim to cosine similarity

Pull Request - State: closed - Opened by Isalia20 8 months ago - 10 comments
Labels: module: nn, triaged, open source, Stale, release notes: onnx, topic: improvements

#148977 - module.cuda() doesn't work under FakeTensorMode

Issue - State: open - Opened by bdhirsh 8 months ago - 4 comments
Labels: module: nn, triaged, oncall: pt2, module: fakeTensor, module: pt2-dispatcher

#148939 - Whether the transposed tensor is contiguous affects the results of the subsequent Linear layer.

Issue - State: closed - Opened by pikerbright 8 months ago - 4 comments
Labels: needs reproduction, module: nn, triaged, module: intel

#148867 - FP16 of weight norm is slower than BF16 on CPU

Issue - State: closed - Opened by jiqing-feng 8 months ago - 1 comment
Labels: module: nn, triaged

#148259 - Raise a warning when `torch.nn.utils.clip_grad_norm_` receives an exhausted generator

Issue - State: closed - Opened by Orimalca 9 months ago - 2 comments
Labels: module: nn, module: error checking, triaged, actionable

#147918 - [FlexAttention] Fix IMA bug

Pull Request - State: closed - Opened by drisspg 9 months ago - 6 comments
Labels: high priority, module: nn, Merged, ciflow/trunk, release notes: nn, bug, module: inductor, ciflow/inductor, merging

#147696 - There is a problem with the wording here.

Issue - State: closed - Opened by sealoongleft 9 months ago - 2 comments
Labels: module: docs, module: nn, triaged

#147522 - Fix the shape check inside gnll loss

Pull Request - State: open - Opened by KohakuBlueleaf 9 months ago - 7 comments
Labels: module: nn, triaged, open source, Stale, topic: not user facing, bug

#147458 - `torch.nn.functional.conv1d` can cause a `Floating point exception (core dumped)`

Issue - State: closed - Opened by cybersupersoap 9 months ago - 2 comments
Labels: module: crash, module: nn, triaged, module: mkldnn, topic: fuzzer

#147209 - Passing `src_key_padding_mask` as `bool` vs `float` causes different outputs from `nn.TransformerEncoderLayer`

Issue - State: closed - Opened by petercall 9 months ago - 3 comments
Labels: module: nn, triaged

#146972 - DISABLED test_per_sample_api_compute_batch_size_not_pytreeable_cpu (main.TestExpandedWeightModuleCPU)

Issue - State: open - Opened by pytorch-bot[bot] 9 months ago - 13 comments
Labels: module: nn, triaged, module: flaky-tests, skipped, oncall: pt2

#146672 - inconsistency in torch.nn.Tanh on CPU and GPU

Issue - State: closed - Opened by alionapi 10 months ago - 1 comment
Labels: module: nn, module: cpu, triaged

#145499 - Remove truncated normal initialization for 16-bit (and lower) tensors

Pull Request - State: closed - Opened by hjlee1371 10 months ago - 4 comments
Labels: module: nn, triaged, open source, Stale

#144245 - F.interpolate returns NAN on MPS if align_corner is True.

Issue - State: closed - Opened by benHeid 11 months ago - 3 comments
Labels: module: nn, triaged, module: correctness (silent), module: mps

#144223 - Support LayerNorm2d

Issue - State: open - Opened by Luciennnnnnn 11 months ago - 2 comments
Labels: module: nn, triaged

#144207 - `pack_padded_sequence ` -> `pad_packed_sequence` can silently truncate input tensor when max length is smaller than actual max sequence length

Issue - State: closed - Opened by brylee10 11 months ago - 2 comments
Labels: module: nn, module: rnn, triaged, actionable

#143758 - Inconsistent results between F.linear and manual computation

Issue - State: closed - Opened by eliahuhorwitz 11 months ago - 1 comment
Labels: module: numerical-stability, module: nn

#143481 - Segmentation fault (core dumped) in `conv3d`

Issue - State: open - Opened by LongZE666 11 months ago - 1 comment
Labels: module: crash, module: nn, module: error checking, module: convolution, triaged, topic: fuzzer

#143480 - Segmentation fault (core dumped) in `conv1d`

Issue - State: open - Opened by LongZE666 11 months ago - 1 comment
Labels: module: crash, module: nn, module: error checking, triaged, module: edge cases, topic: fuzzer

#141315 - GroupNorm Doc Typo: Should Be Variance

Issue - State: closed - Opened by YifanShenSZ 12 months ago
Labels: module: docs, module: nn, triaged, module: norms and normalization

#140986 - Improves wording and grammar on the documentation for nn/module.py

Issue - State: closed - Opened by fmgblackwolf about 1 year ago - 1 comment
Labels: module: docs, module: nn, triaged

#140934 - When I upgraded torch to 2.3.1, [F.conv2d] performance got really bad!

Issue - State: closed - Opened by MrXsc about 1 year ago - 2 comments
Labels: needs reproduction, module: performance, module: nn, module: convolution, triaged

#140928 - The RNN implementation example in the documentation may be incorrect

Issue - State: closed - Opened by suzakuwcx about 1 year ago - 1 comment
Labels: module: nn, module: rnn, triaged, topic: docs

#140923 - Segmentation fault (core dumped) in `max_pool2d_with_indices_backward`

Issue - State: closed - Opened by LongZE666 about 1 year ago - 5 comments
Labels: module: crash, module: nn, module: error checking, triaged, module: edge cases

#140215 - Fix RMSNorm Notation: Parentheses, Indices, Comma

Pull Request - State: open - Opened by d-kleine about 1 year ago - 4 comments
Labels: module: nn, open source, topic: docs, topic: not user facing

#140215 - Fix RMSNorm Notation: Parentheses, Indices, Comma

Pull Request - State: closed - Opened by d-kleine about 1 year ago - 11 comments
Labels: module: nn, triaged, open source, Merged, ciflow/trunk, release notes: nn, topic: docs, topic: not user facing, merging

#140165 - Doc issue: Fix Inconsistent Notation for RMSNorm

Issue - State: closed - Opened by d-kleine about 1 year ago - 1 comment
Labels: module: docs, module: nn, triaged

#139467 - Seperate grad norm computation from `torch.nn.utils.clip_grad_norm_`

Issue - State: closed - Opened by zijian-hu about 1 year ago - 5 comments
Labels: module: nn, triaged

#139387 - Add prepacking for linear weights

Pull Request - State: closed - Opened by Ryo-not-rio about 1 year ago - 31 comments
Labels: oncall: distributed, module: nn, module: cpu, triaged, module: mkldnn, open source, module: amp (automated mixed precision), module: arm, NNC, release notes: quantization, release notes: linalg_frontend, module: inductor, module: dynamo, module: compiled autograd

#138401 - Documentation for torch.nn.RNN contains an INACCURATE code snippet

Issue - State: closed - Opened by Datarian723 about 1 year ago - 2 comments
Labels: module: docs, module: nn, module: rnn, triaged, actionable

#138385 - `torch.nn.functional.softshrink` returns 0 on `NaN` input.

Issue - State: closed - Opened by dlibk about 1 year ago
Labels: module: nn, triaged, module: edge cases

#138025 - RMSNorm with bias kwarg option

Issue - State: closed - Opened by PeaBrane about 1 year ago - 1 comment
Labels: module: nn, triaged

#137710 - Updating a module parameter in-place with a different shape does not update the tensor shape in `state_dict`

Issue - State: closed - Opened by fxmarty-amd about 1 year ago - 4 comments
Labels: module: nn, triaged, module: viewing and reshaping

#136925 - The doc of `nn.RNN()` should say there is `hx` parameter

Issue - State: closed - Opened by hyperkai about 1 year ago - 1 comment
Labels: module: nn, module: rnn, triaged

#136597 - [Doc issue] RMSNorm formula

Issue - State: closed - Opened by d-kleine about 1 year ago
Labels: module: docs, module: nn, triaged

#136409 - Linear Sum Assignment

Issue - State: open - Opened by adossantos21 about 1 year ago
Labels: module: nn, module: optimizer, triaged

#135880 - The order of the parameters of `nn.Conv1d()`, `nn.Conv2d()` and `nn.Conv3d()` should be explained in the actual order of the parameters.

Issue - State: closed - Opened by hyperkai about 1 year ago - 3 comments
Labels: module: docs, module: nn, good first issue, triaged, actionable

#135596 - Validate input types for `torch.nn.Linear` and `torch.nn.Bilinear`

Pull Request - State: closed - Opened by sanskarmodi8 about 1 year ago - 17 comments
Labels: module: nn, module: error checking, triaged, open source, Merged, ciflow/trunk, release notes: python_frontend, topic: improvements

#135596 - Validate input types for `torch.nn.Linear` and `torch.nn.Bilinear`

Pull Request - State: closed - Opened by sanskarmodi8 about 1 year ago - 24 comments
Labels: module: nn, module: error checking, triaged, open source, Merged, Reverted, Stale, ciflow/trunk, release notes: python_frontend, topic: improvements

#135519 - torch.nn.ReplicationPad and torch.nn.ReflectionPad does not support float16

Issue - State: closed - Opened by jiren-the-gray about 1 year ago - 2 comments
Labels: module: nn, triaged, actionable

#135463 - Setting a wrong type value instead of a correct integer typed value to `in_features` or `out_features` argument of `nn.Linear()` gets indirect error messages

Issue - State: closed - Opened by hyperkai about 1 year ago - 6 comments
Labels: module: nn, good first issue, module: error checking, triaged, module: python frontend, module: edge cases

#134760 - How to correctly release the memory of a tensor

Issue - State: closed - Opened by huangqiaobo about 1 year ago - 1 comment
Labels: module: nn, module: memory usage, triaged

#134643 - Allow optional positional arguments for `torch.func.functional_call`

Pull Request - State: closed - Opened by why-in-Shanghaitech about 1 year ago - 12 comments
Labels: module: nn, triaged, open source, Merged, ciflow/trunk, release notes: nn, merging, suppress-bc-linter

#134643 - Allow optional positional arguments for `torch.func.functional_call`

Pull Request - State: closed - Opened by why-in-Shanghaitech about 1 year ago - 12 comments
Labels: module: nn, triaged, open source, Merged, ciflow/trunk, release notes: nn, suppress-bc-linter