-
Notifications
You must be signed in to change notification settings - Fork 28.9k
All issues
Issue creation is restricted in this repository
Issues
is:issue state:open
is:issue state:open
Search results
[inductor] torch.compile crashes on view(dtype).clone().copy_() chains: "too many values to unpack" in generated wrapper (arg-count mismatch), CPU and CUDA
bot-triagedThis is a label only to be used by the auto triage botThis is a label only to be used by the auto triage bottriagedThis issue has been looked at a team member, and triaged and prioritized into an appropriate moduleThis issue has been looked at a team member, and triaged and prioritized into an appropriate moduleStatus: Open.#193705 In pytorch/pytorch;[ROCm]
nn.Embedding(sparse=True)backward computes ~10^5x incorrect gradient magnitude on ROCm (gfx1201 / RDNA4)bot-triagedThis is a label only to be used by the auto triage botThis is a label only to be used by the auto triage botmodule: autogradRelated to torch.autograd, and the autograd engine in generalRelated to torch.autograd, and the autograd engine in generalmodule: correctness (silent)issue that returns an incorrect result silentlyissue that returns an incorrect result silentlymodule: rocmAMD GPU support for PytorchAMD GPU support for Pytorchmodule: sparseRelated to torch.sparseRelated to torch.sparseStatus: Open.#193704 In pytorch/pytorch;[Inductor] User-defined Triton access analysis is exponential on shared TTIR DAGs
bot-triagedThis is a label only to be used by the auto triage botThis is a label only to be used by the auto triage botmodule: compile-timeCompilation mechanism or time spent in (re)compilation, tracing, startupCompilation mechanism or time spent in (re)compilation, tracing, startupmodule: regressionIt used to work, and now it doesn'tIt used to work, and now it doesn'tmodule: user tritonrelated to ability to directly torch.compile triton kernelsrelated to ability to directly torch.compile triton kernelsStatus: Open.#193703 In pytorch/pytorch;[CUDA] illegal memory read in batch_norm_update_stats_and_invert with different running_mean and running_var dtypes
bot-triagedThis is a label only to be used by the auto triage botThis is a label only to be used by the auto triage botmodule: crashProblem manifests as a hard crash, as opposed to a RuntimeErrorProblem manifests as a hard crash, as opposed to a RuntimeErrormodule: cudaRelated to torch.cuda, and CUDA support in generalRelated to torch.cuda, and CUDA support in generalStatus: Open.#193698 In pytorch/pytorch;[CUDA] illegal memory atomic on atomicadaptivemaxgradinput
bot-triagedThis is a label only to be used by the auto triage botThis is a label only to be used by the auto triage botmodule: crashProblem manifests as a hard crash, as opposed to a RuntimeErrorProblem manifests as a hard crash, as opposed to a RuntimeErrormodule: cudaRelated to torch.cuda, and CUDA support in generalRelated to torch.cuda, and CUDA support in generalStatus: Open.#193696 In pytorch/pytorch;[CUDA] illegal memory read on reflection_pad2d_backward_det_out_kernel
bot-triagedThis is a label only to be used by the auto triage botThis is a label only to be used by the auto triage botmodule: crashProblem manifests as a hard crash, as opposed to a RuntimeErrorProblem manifests as a hard crash, as opposed to a RuntimeErrormodule: cudaRelated to torch.cuda, and CUDA support in generalRelated to torch.cuda, and CUDA support in generalStatus: Open.#193694 In pytorch/pytorch;- Status: Open.#193693 In pytorch/pytorch;
[CUDA] illegal memory read on compute_grad_weight_bags
bot-triagedThis is a label only to be used by the auto triage botThis is a label only to be used by the auto triage botmodule: crashProblem manifests as a hard crash, as opposed to a RuntimeErrorProblem manifests as a hard crash, as opposed to a RuntimeErrormodule: cudaRelated to torch.cuda, and CUDA support in generalRelated to torch.cuda, and CUDA support in generalStatus: Open.#193691 In pytorch/pytorch;Distribution.cdf rejects values outside the support, though the CDF is defined on all of R (and the implementations already clamp for it)
bot-triagedThis is a label only to be used by the auto triage botThis is a label only to be used by the auto triage botmodule: distributionsRelated to torch.distributionsRelated to torch.distributionstriagedThis issue has been looked at a team member, and triaged and prioritized into an appropriate moduleThis issue has been looked at a team member, and triaged and prioritized into an appropriate moduleStatus: Open.#193690 In pytorch/pytorch;[CUDA] 32-bit input overflow causes OOB read in upsample_bilinear2d_out_frame
bot-triagedThis is a label only to be used by the auto triage botThis is a label only to be used by the auto triage botmodule: 64-bitProblems related to incorrectly using 32-bit integers when 64-bit is needed (e.g., 8G tensors)Problems related to incorrectly using 32-bit integers when 64-bit is needed (e.g., 8G tensors)module: cudaRelated to torch.cuda, and CUDA support in generalRelated to torch.cuda, and CUDA support in generalStatus: Open.#193689 In pytorch/pytorch;LinearLR with total_iters=0 applies start_factor permanently instead of acting as a no-op
bot-triagedThis is a label only to be used by the auto triage botThis is a label only to be used by the auto triage botmodule: correctness (silent)issue that returns an incorrect result silentlyissue that returns an incorrect result silentlyStatus: Open.#193682 In pytorch/pytorch;Silent wrong gradients: nn.RReLU under non-reentrant checkpointing saves uninitialized noise
bot-triagedThis is a label only to be used by the auto triage botThis is a label only to be used by the auto triage botmodule: activation checkpointingRelated to activation checkpointingRelated to activation checkpointingmodule: autogradRelated to torch.autograd, and the autograd engine in generalRelated to torch.autograd, and the autograd engine in generalmodule: correctness (silent)issue that returns an incorrect result silentlyissue that returns an incorrect result silentlyStatus: Open.#193671 In pytorch/pytorch;