pytorch/pytorch issues and pull requests

#161915 - `share_memory_` support for `NestedTensor`

Issue - State: closed - Opened by Phimos 3 months ago - 1 comment
Labels: triaged, module: nestedtensor

#161734 - [BC Breaking] Remove flex + njt code paths

Pull Request - State: closed - Opened by drisspg 3 months ago - 7 comments
Labels: module: nestedtensor, Merged, ciflow/trunk, release notes: nn, topic: bc breaking, merging, suppress-bc-linter, module: flex attention

#161434 - [cuDNN][SDPA][Nested Tensor] add forward/backward caching support for cuDNN SDPA Nested tensor/varlen

Pull Request - State: open - Opened by eqy 3 months ago
Labels: module: cudnn, module: cuda, open source, module: nestedtensor, topic: not user facing, module: sdpa

#157004 - DISABLED test_scaled_dot_product_attention_input_dim_4_cuda (main.TestNestedTensorDeviceTypeCUDA)

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

#156660 - Adds support for Nested Jagged Tensor in Multihead Attention

Pull Request - State: open - Opened by rusdes 5 months ago - 3 comments
Labels: open source, module: nestedtensor

#153237 - [NJT] NestedTensor repr has contiguous=True while the NJT isn't contiguous

Issue - State: closed - Opened by mdeff 6 months ago - 2 comments
Labels: triaged, module: nestedtensor, actionable

#152950 - Add NestedTensorHPU to to_padded_tensor in native_functions.yaml

Pull Request - State: open - Opened by sfraczek 7 months ago - 8 comments
Labels: triaged, open source, module: nestedtensor, ciflow/trunk, topic: improvements, release notes: nested tensor, ciflow/xpu, release notes: xpu

#152235 - [CUDA][SDPA] bump fudge factor in `test_sdpa` in `test_nestedtensor`

Pull Request - State: closed - Opened by eqy 7 months ago - 6 comments
Labels: module: cuda, open source, module: nestedtensor, Merged, ciflow/trunk, topic: not user facing, merging, module: sdpa

#152068 - Compiling attention (SDPA) with nested tensors fails when using DDP

Issue - State: open - Opened by mahyarkoy 7 months ago - 2 comments
Labels: oncall: distributed, triaged, module: nestedtensor, oncall: pt2, module: sdpa

#150172 - Relax njt x njt to dense matmul reduction checks

Pull Request - State: open - Opened by JCBrouwer 8 months ago - 7 comments
Labels: triaged, open source, module: nestedtensor

#149892 - [draft] Add support in Flex for non-contiguous NJT

Pull Request - State: closed - Opened by ani300 8 months ago - 2 comments
Labels: open source, module: nestedtensor, Stale, release notes: nested tensor

#149022 - Support broadcast for nested tensors

Issue - State: closed - Opened by shadow150519 8 months ago - 2 comments
Labels: triaged, module: nestedtensor

#148988 - second derivative of scaled_dot_product_attention does not work for nested tensors

Issue - State: closed - Opened by mahyarkoy 8 months ago - 2 comments
Labels: module: autograd, triaged, module: nestedtensor, actionable

#147032 - [NJT] fix flop counter for SDPA & test

Pull Request - State: closed - Opened by davidberard98 9 months ago - 4 comments
Labels: module: nestedtensor, Merged, ciflow/trunk, topic: bug fixes, release notes: nested tensor, merging

#146932 - Use of @property on in-graph constructed NJT fails Dynamo tracing

Issue - State: closed - Opened by jbschlosser 9 months ago
Labels: triaged, module: nestedtensor, oncall: pt2, module: dynamic shapes, module: dynamo, dynamo-triage-jan2025

#146404 - Can't back prop through NJT matrix multiplication when output is strided tensor

Issue - State: closed - Opened by michael-diggin 10 months ago
Labels: triaged, module: nestedtensor, actionable

#145402 - Update OSS nested tensor docs to focus on NJT

Pull Request - State: closed - Opened by jbschlosser 10 months ago - 17 comments
Labels: module: nestedtensor, Merged, ciflow/trunk, topic: docs, release notes: nested tensor, merging

#145257 - nn.Embedding backwards pass for nested tensors

Issue - State: closed - Opened by kkj15dk 10 months ago - 3 comments
Labels: triaged, module: nestedtensor

#145256 - loss.backward() breaking somewhere when modulating a nested tensor using scale and shift (RuntimeError: Function AddBackward0 returned an invalid gradient at index 0 - got [1, 4, 64] but expected shape compatible with [4, 1, 64])

Issue - State: closed - Opened by kkj15dk 10 months ago - 2 comments
Labels: module: autograd, triaged, module: nestedtensor

#145214 - Nested tensor support for pointwise matrix multiplication of nested tensor and normal tensor

Issue - State: open - Opened by kkj15dk 10 months ago - 4 comments
Labels: triaged, module: nestedtensor

#144936 - DISABLED test_compile_forward_clone_cpu_float32 (main.TestNestedTensorOpInfoCPU)

Issue - State: closed - Opened by pytorch-bot[bot] 10 months ago - 3 comments
Labels: triaged, module: flaky-tests, module: nestedtensor, skipped

#144935 - DISABLED test_compile_forward_select_cuda_float32 (main.TestNestedTensorOpInfoCUDA)

Issue - State: open - Opened by pytorch-bot[bot] 10 months ago
Labels: triaged, module: flaky-tests, module: nestedtensor, skipped

#144761 - torch.rand_like() for nested tensors

Issue - State: closed - Opened by kkj15dk 10 months ago - 1 comment
Labels: triaged, module: nestedtensor

#144587 - Implement backward for NJT matmul

Pull Request - State: closed - Opened by jbschlosser 10 months ago - 9 comments
Labels: module: nestedtensor, Merged, ciflow/trunk, topic: bug fixes, release notes: nested tensor, merging

#143788 - Flex attention with nested tensors, bug in `create_nested_block_mask`

Issue - State: closed - Opened by VivekPanyam 11 months ago - 2 comments
Labels: triaged, module: nestedtensor, oncall: pt2, module: higher order operators, module: pt2-dispatcher, module: flex attention

#143026 - DISABLED test_backward_atan2_cuda_float32 (main.TestNestedTensorOpInfoCUDA)

Issue - State: open - Opened by pytorch-bot[bot] 11 months ago
Labels: triaged, module: flaky-tests, module: nestedtensor, skipped

#142993 - DISABLED test_compile_forward_special_erfcx_cuda_float32 (main.TestNestedTensorOpInfoCUDA)

Pull Request - State: open - Opened by pytorch-bot[bot] 11 months ago
Labels: triaged, module: flaky-tests, module: nestedtensor, skipped

#142913 - DISABLED test_backward_tanh_cuda_float32 (main.TestNestedTensorOpInfoCUDA)

Issue - State: open - Opened by pytorch-bot[bot] 11 months ago
Labels: triaged, module: flaky-tests, module: nestedtensor, skipped

#142662 - DISABLED test_backward_polygamma_polygamma_n_0_cuda_float32 (main.TestNestedTensorOpInfoCUDA)

Issue - State: open - Opened by pytorch-bot[bot] 11 months ago
Labels: triaged, module: flaky-tests, module: nestedtensor, skipped

#142618 - DISABLED test_compile_forward_any_cuda_float32 (main.TestNestedTensorOpInfoCUDA)

Issue - State: open - Opened by pytorch-bot[bot] 11 months ago
Labels: triaged, module: flaky-tests, module: nestedtensor, skipped

#142614 - DISABLED test_backward_rpow_cuda_float32 (main.TestNestedTensorOpInfoCUDA)

Issue - State: open - Opened by pytorch-bot[bot] 11 months ago
Labels: triaged, module: flaky-tests, module: nestedtensor, skipped

#142561 - DISABLED test_forward_chalf_cuda_float32 (main.TestNestedTensorOpInfoCUDA)

Issue - State: open - Opened by pytorch-bot[bot] 11 months ago
Labels: triaged, module: flaky-tests, module: nestedtensor, skipped

#142552 - DISABLED test_compile_forward_min_reduction_with_dim_cuda_float32 (main.TestNestedTensorOpInfoCUDA)

Issue - State: open - Opened by pytorch-bot[bot] 11 months ago
Labels: triaged, module: flaky-tests, module: nestedtensor, skipped

#142157 - Some NJT tests are not failing in CI as expected

Issue - State: closed - Opened by jbschlosser 12 months ago - 10 comments
Labels: module: ci, module: tests, triaged, module: nestedtensor

#141486 - FlexAttention with compiled block mask is slow when varying sequence lengths

Issue - State: closed - Opened by samvanstroud 12 months ago - 5 comments
Labels: triaged, module: nestedtensor, oncall: pt2, module: higher order operators, module: pt2-dispatcher, module: flex attention

#141380 - Add bfloat16 support to torch.bmm(NST, NST)

Pull Request - State: closed - Opened by cpuhrsch 12 months ago - 3 comments
Labels: triaged, module: nestedtensor, Merged, ciflow/trunk, topic: improvements, release notes: nested tensor, merging

#141306 - Issue with torch.load failing to resolve torch.nested._internal during checkpoint loading (Nightly PyTorch ROCm 6.2)

Issue - State: closed - Opened by Mahdi-CV 12 months ago - 4 comments
Labels: module: rocm, module: serialization, triaged, module: nestedtensor

#141292 - DISABLED test_linear_backward_memory_usage_cuda_float32 (main.TestNestedTensorSubclassCUDA)

Issue - State: open - Opened by pytorch-bot[bot] 12 months ago - 4 comments
Labels: triaged, module: flaky-tests, module: nestedtensor, skipped

#141178 - [cuDNN][SDPA][Nested Tensor] Experimental cuDNN Nested Tensor SDPA Support (forward only)

Pull Request - State: closed - Opened by eqy 12 months ago - 39 comments
Labels: module: cudnn, triaged, open source, module: nestedtensor, Merged, Reverted, ciflow/binaries, ciflow/trunk, topic: not user facing, merging, ci-no-td, module: sdpa

#141112 - Memory explodes when applying Linear to reshaped nested tensor

Issue - State: closed - Opened by mahyarkoy 12 months ago - 1 comment
Labels: high priority, triage review, triaged, module: nestedtensor