data/issues.html

<style>

:root { color-scheme: light; }

.bi-wrap { font-family: Inter, ui-sans-serif, system-ui, -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif; color: #1f2328 !important; }

.bi-wrap h1, .bi-wrap h2, .bi-wrap h3, .bi-wrap p, .bi-wrap strong, .bi-wrap td, .bi-wrap th, .bi-wrap div { color: #1f2328 !important; }

.bi-band { border: 1px solid #d8dee4; border-radius: 8px; background: #ffffff !important; padding: 16px; margin: 0 0 12px; }

.bi-title { display: flex; gap: 12px; align-items: flex-start; justify-content: space-between; margin-bottom: 12px; }

.bi-title h1, .bi-title h2 { margin: 0; line-height: 1.15; letter-spacing: 0; }

.bi-title h1 { font-size: 1.55rem; }

.bi-title h2 { font-size: 1.12rem; }

.bi-title p { margin: 4px 0 0; color: #57606a; max-width: 820px; }

.status-pill { border-radius: 999px; padding: 5px 9px; font-size: .78rem; font-weight: 700; background: #dafbe1; color: #116329 !important; white-space: nowrap; }

.metric-grid { display: grid; grid-template-columns: repeat(auto-fit, minmax(138px, 1fr)); gap: 8px; }

.metric { border: 1px solid #d0d7de; border-radius: 8px; padding: 10px; background: #f6f8fa !important; min-height: 86px; }

.metric-value { font-size: 1.45rem; font-weight: 800; line-height: 1.1; color: #0969da !important; }

.metric-label { margin-top: 4px; font-size: .84rem; font-weight: 700; }

.metric-note { margin-top: 2px; font-size: .76rem; color: #57606a !important; }

.callout-grid { display: grid; grid-template-columns: repeat(auto-fit, minmax(220px, 1fr)); gap: 8px; }

.callout { border-left: 4px solid #0969da; background: #f6f8fa !important; border-radius: 8px; padding: 10px 12px; }

.callout:nth-child(2) { border-left-color: #1a7f37; }

.callout:nth-child(3) { border-left-color: #bf8700; }

.callout:nth-child(4) { border-left-color: #cf222e; }

.callout strong { display: block; margin-bottom: 4px; }

.group-grid { display: grid; grid-template-columns: repeat(auto-fit, minmax(310px, 1fr)); gap: 10px; }

.group-card { border: 1px solid #d8dee4; border-radius: 8px; padding: 12px; background: #fff !important; }

.group-card h3 { margin: 0 0 7px; font-size: 1rem; line-height: 1.25; letter-spacing: 0; }

.meta-row { display: flex; flex-wrap: wrap; gap: 6px; margin: 6px 0; }

.chip-row { display: inline-flex; flex-wrap: wrap; gap: 5px; align-items: center; vertical-align: middle; }

.chip { display: inline-flex; align-items: center; border: 1px solid #d0d7de; border-radius: 999px; padding: 2px 7px; font-size: .78rem; background: #f6f8fa !important; color: #24292f !important; }

.chip.bucket { border-color: #9a6700; background: #fff8c5 !important; }

.chip.kind { border-color: #0969da; background: #ddf4ff !important; }

.workflow-grid { display: grid; grid-template-columns: repeat(auto-fit, minmax(260px, 1fr)); gap: 10px; }

.workflow-card { border: 1px solid #d8dee4; border-radius: 8px; padding: 12px; background: #fff !important; }

.workflow-card h3 { margin: 0 0 7px; font-size: 1rem; line-height: 1.25; letter-spacing: 0; }

.workflow-card .move-count { font-weight: 800; color: #0969da !important; }

.move-list { display: grid; gap: 10px; }

.move-row { border: 1px solid #d8dee4; border-radius: 8px; background: #fff !important; }

.move-row summary { list-style: none; cursor: pointer; padding: 12px; }

.move-row summary::-webkit-details-marker { display: none; }

.move-summary { display: grid; grid-template-columns: minmax(190px, 1fr) minmax(260px, 1.45fr) minmax(260px, 1.45fr) minmax(150px, .8fr) minmax(250px, 1.35fr) auto; gap: 16px; align-items: start; }

.move-title, .move-field { min-width: 0; }

.move-title h3 { margin: 0 0 4px; font-size: 1rem; line-height: 1.25; letter-spacing: 0; }

.move-field span { display: block; color: #57606a !important; font-size: .72rem; font-weight: 800; text-transform: uppercase; letter-spacing: 0; }

.move-field p { margin: 3px 0 0; font-size: .84rem; line-height: 1.35; overflow-wrap: anywhere; }

.move-field .chip, .move-title { overflow-wrap: anywhere; }

.view-bundles { align-self: center; justify-self: end; border: 1px solid #000000; border-radius: 999px; color: #000000 !important; background: #ffffff !important; padding: 4px 9px; font-size: .78rem; font-weight: 800; white-space: nowrap; }

.move-row[open] .view-bundles { background: #f6f8fa !important; }

.move-expanded { border-top: 1px solid #d8dee4; padding: 0 12px 12px; }

.bundle-row { border-top: 1px solid #eef1f4; padding: 12px 0; }

.bundle-row:first-child { border-top: none; }

.bundle-row h4 { margin: 0 0 6px; font-size: .94rem; line-height: 1.25; letter-spacing: 0; }

.bundle-scope { display: grid; grid-template-columns: repeat(auto-fit, minmax(220px, 1fr)); gap: 7px; margin-top: 8px; }

.evidence-list { margin: 8px 0 0 18px; padding: 0; color: #57606a !important; font-size: .82rem; line-height: 1.35; }

.evidence-list li { margin: 3px 0; }

.muted { color: #57606a !important; }

.evidence { font-size: .84rem; color: #57606a !important; }

.scroll-table { overflow-x: auto; border: 1px solid #d8dee4; border-radius: 8px; background: #fff !important; }

table.bi-table { width: 100%; border-collapse: collapse; font-size: .86rem; }

.bi-table th { text-align: left; background: #f6f8fa !important; color: #57606a !important; font-size: .78rem; border-bottom: 1px solid #d8dee4; padding: 8px; white-space: nowrap; }

.bi-table td { border-bottom: 1px solid #eef1f4; padding: 8px; vertical-align: top; min-width: 86px; }

.bi-table tr:last-child td { border-bottom: none; }

.wide { min-width: 280px; }

.tight { max-width: 180px; }

a { color: #0969da !important; text-decoration: none; }

a:hover { text-decoration: underline; }

.bi-wrap, .bi-wrap * { color: #000000 !important; }

@media (max-width: 720px) {

  .bi-title { display: block; }

  .status-pill { display: inline-block; margin-top: 8px; }

  .group-grid { grid-template-columns: 1fr; }

  .move-summary { grid-template-columns: 1fr; }

  .view-bundles { justify-self: start; }

}

</style>
<div class='bi-wrap'><section class='bi-band'><div class='bi-title'><div><h2>Open Issues</h2><p>Public-safe normalized rows for auditability.</p></div></div><div class='scroll-table'><table class='bi-table'><thead><tr><th>Issue</th><th>Bucket</th><th>Score</th><th>Linked</th><th>Labels</th><th>Excerpt</th></tr></thead><tbody><tr><td><a href='https://github.com/pytorch/pytorch/issues/180948' target='_blank' rel='noopener noreferrer'>Issue #180948: DISABLED test_emulate_precision_casts_preserves_explicit_precision_cast_float16 (__main__.CudaReproTests)</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>module: tests, triaged, skipped, oncall: pt2, module: inductor, module: xpu, bot-triaged</td><td><div class='wide'>Platforms: xpu This test was disabled because it is failing on main branch (recent examples). cc @mruberry @chauhang @penguinwu @voznesenskym @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @ipiszy @kadeng @muchulee8 @amjames @aakhundov @coconutruben @jataylo @gujinghui @fe...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189322' target='_blank' rel='noopener noreferrer'>Issue #189322: DISABLED test_aoti_debug_printer_codegen_cuda (__main__.AOTInductorTestDualWrapper)</a></td><td>needs design decision</td><td>65</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>module: flaky-tests, skipped, oncall: pt2, module: inductor</td><td><div class='wide'>Platforms: inductor This test was disabled because it is failing in CI. See recent examples and the most recent trunk workflow logs. Over the past 6 hours, it has been determined flaky in 3 workflow(s) with 3 failures and 3 successes. Debugging instructions (after clicking on the recent samples link): DO NOT ASSUME...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189323' target='_blank' rel='noopener noreferrer'>Issue #189323: DISABLED test_fake_export_contiguous_cpu_float32 (__main__.TestExportOpInfoCPU)</a></td><td>needs design decision</td><td>65</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>module: flaky-tests, skipped, oncall: pt2, oncall: export</td><td><div class='wide'>Platforms: linux This test was disabled because it is failing in CI. See recent examples and the most recent trunk workflow logs. Over the past 6 hours, it has been determined flaky in 7 workflow(s) with 7 failures and 7 successes. Debugging instructions (after clicking on the recent samples link): DO NOT ASSUME THI...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/119537' target='_blank' rel='noopener noreferrer'>Issue #119537: DISABLED test_source_multithreaded_open_in_scope_work_in_main_thread_False (__main__.TestProfiler)</a></td><td>needs design decision</td><td>65</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>module: flaky-tests, skipped, oncall: profiler</td><td><div class='wide'>Platforms: dynamo, linux, mac, macos, rocm, slow, asan This test was disabled because it is failing in CI. See recent examples and the most recent trunk workflow logs. Over the past 3 hours, it has been determined flaky in 4 workflow(s) with 12 failures and 4 successes. Debugging instructions (after clicking on the...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/119576' target='_blank' rel='noopener noreferrer'>Issue #119576: DISABLED test_source_multithreaded_multiple_preexisting_work_in_main_thread_True (__main__.TestProfiler)</a></td><td>needs design decision</td><td>65</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>module: flaky-tests, skipped, oncall: profiler</td><td><div class='wide'>Platforms: linux, mac, macos, rocm, asan, slow This test was disabled because it is failing in CI. See recent examples and the most recent trunk workflow logs. Over the past 3 hours, it has been determined flaky in 7 workflow(s) with 21 failures and 7 successes. Debugging instructions (after clicking on the recent s...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/180194' target='_blank' rel='noopener noreferrer'>Issue #180194: DISABLED test_vjp_grad_cpu (__main__.TestComposabilityCPU)</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>triaged, module: flaky-tests, skipped, oncall: pt2, module: functorch</td><td><div class='wide'>Platforms: dynamo This test was disabled because it is failing in CI. See recent examples and the most recent trunk workflow logs. Over the past 6 hours, it has been determined flaky in 3 workflow(s) with 3 failures and 3 successes. Debugging instructions (after clicking on the recent samples link): DO NOT ASSUME TH...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/174507' target='_blank' rel='noopener noreferrer'>Issue #174507: PyTorch Stable ABI support</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>feature, module: abi, triaged, bot-triaged</td><td><div class='wide'>🐛 Describe the bug Hello @janeyx99 I am one of owners of https://github.com/amd/quark and I am super excited in converting our runtime-compiled Torch C++ extensions into pre-compiled libraries shipped with Quark using the PyTorch&#x27;s stable ABI When I checked our code that uses torch C++, I have identified APIs that i...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189164' target='_blank' rel='noopener noreferrer'>Issue #189164: [CI][B200] B200 Runner Maintenance Notification</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>module: ci, triaged, module: infra, bot-triaged</td><td><div class='wide'>We are performing B200 runner maintenance one by one for the six runners. Jobs may encounter failures during this time and may need to be restarted. cc @malfet @pytorch/pytorch-dev-infra @huydhn @atalman @ptrblck @eqy @tinglvv</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189194' target='_blank' rel='noopener noreferrer'>Issue #189194: [ROCm] torch 2.13 wheel: &quot;Can&#x27;t detect vectorized ISA for CPU&quot; in torch.compile smoke test on non-ROCm image (regression vs 2.12.1)</a></td><td>needs design decision</td><td>115</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>high priority, triage review, module: binaries, module: rocm, triaged, module: vectorization, module: regression, oncall: pt2, module: inductor, bot-triaged</td><td><div class='wide'>🐛 Describe the bug The torch==2.13.0+rocm7.2 wheel fails the torch.compile smoke test with RuntimeError: Can&#x27;t detect vectorized ISA for CPU when it is imported/run on a non-ROCm image (e.g. pytorch/almalinux-builder:cpu-main), where torch.cuda.is_available() is False and the compile check exercises the CPU / Induct...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189308' target='_blank' rel='noopener noreferrer'>Issue #189308: `torch.distributed.checkpoint`: `.metadata` loaded with raw `pickle.load` on the public `load()` path — same class you already warn about on `torch.load`</a></td><td>needs design decision</td><td>65</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>oncall: distributed, oncall: distributed checkpointing, bot-triaged</td><td><div class='wide'>🐛 Describe the bug I first reported this privately as a security advisory (GHSA-ppvg-47fv-5f4w). It was closed under the SECURITY.md policy. In that thread I asked whether it might still be worth tracking as a documentation/API-hardening issue and whether you would prefer that I open a regular issue; I did not recei...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189309' target='_blank' rel='noopener noreferrer'>Issue #189309: [Stable ABI] Add a has_side_effects op tag. DCE silently deletes side-effect-only op calls</a></td><td>needs design decision</td><td>65</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>oncall: jit, bot-triaged</td><td><div class='wide'>🚀 The feature, motivation and pitch While porting torchvision to stable ABI, I noticed that ops that return nothing and exist only for their side effects cannot be registered safely. StableLibrary::def only takes a schema string so the op gets FROM_SCHEMA alias analysis, and TorchScript DCE deletes calls to an op wi...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/150199' target='_blank' rel='noopener noreferrer'>Issue #150199: `asarray`: device does not propagate from input to output after `set_default_device`</a></td><td>has linked PR</td><td>44</td><td><span class='chip-row'><span class='chip'>PR #188795</span></span></td><td>triaged, module: python array api, module: python frontend</td><td><div class='wide'>🐛 Describe the bug The documentation of asarray states: device (torch.device, optional) – the device of the returned tensor. Default: None, which causes the device of obj to be used. Or, if obj is a Python sequence, the current default device will be used. The described behaviour is coherent with the specification o...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/141287' target='_blank' rel='noopener noreferrer'>Issue #141287: MPS operator coverage tracking issue (2.6+ version)</a></td><td>has linked PR</td><td>44</td><td><span class='chip-row'><span class='chip'>PR #178031</span></span></td><td>feature, triaged, tracker, module: mps</td><td><div class='wide'>🐛 Describe the bug This issue is to have a centralized place to list and track work on adding support to new ops for the MPS backend. Most requested ops (extracted from comments to this issue and #77764) PyTorch MPS Ops Project : Project to track all the ops for MPS backend. There are a very large number of operator...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189298' target='_blank' rel='noopener noreferrer'>Issue #189298: torch.jit.trace fails with MaxUnpool in 2.13</a></td><td>needs design decision</td><td>65</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>oncall: jit, bot-triaged</td><td><div class='wide'>🐛 Describe the bug Reproducible Example import torch def main(): pool = torch.nn.MaxPool2d(2, return_indices=True) out, indices = pool(torch.randn(1, 10, 10, 10)) unpool = torch.nn.MaxUnpool2d(2) _ = torch.jit.trace(unpool, (out, indices)) if __name__ == &quot;__main__&quot;: main() error message: TypeError: cond must be a bo...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189293' target='_blank' rel='noopener noreferrer'>Issue #189293: AttributeError: Can&#x27;t get local object &#x27;WeakValueDictionary.__init__.&lt;locals&gt;.remove&#x27; during AOTAutograd graph caching on Python 3.13</a></td><td>needs design decision</td><td>65</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>triage review, module: pickle, tensor subclass, oncall: pt2, module: aotdispatch, module: pt2-dispatcher, module: python version, bot-triaged</td><td><div class='wide'>🐛 Describe the bug When utilizing torch.compile with graph caching enabled (TORCHINDUCTOR_FX_GRAPH_CACHE=1) on Python 3.13 and PyTorch 2.14 Nightly, the compilation disk cache subsystem crashes during the serialization phase of tensor subclass metadata. Specifically, when a custom tensor subclass (such as an advance...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189150' target='_blank' rel='noopener noreferrer'>Issue #189150: SIGSEGV / cudaErrorIllegalAddress when replaying a CUDA graph that captured multiple training iterations of two trainers over deeply unrolled recurrent modules</a></td><td>needs design decision</td><td>115</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>triage review, module: crash, module: cuda, module: cuda graphs, bot-triaged</td><td><div class='wide'>🐛 Describe the bug Capturing N training iterations (forward + backward + AdamW(capturable=True).step()) of TWO independent trainers into one torch.cuda.CUDAGraph produces a graph whose replay crashes. In the script below each trainer owns M=4 plain nn.Modules called in a Python loop - no functorch anywhere - and the...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/188898' target='_blank' rel='noopener noreferrer'>Issue #188898: DISABLED test_vmap_over_vmap_two_inputs (__main__.FuncTorchHigherOrderOpTestsWithCompiledAutograd)</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>triaged, module: flaky-tests, skipped, oncall: pt2, module: inductor</td><td><div class='wide'>Platforms: linux, rocm, slow This test was disabled because it is failing in CI. See recent examples and the most recent trunk workflow logs. Over the past 6 hours, it has been determined flaky in 3 workflow(s) with 6 failures and 3 successes. Debugging instructions (after clicking on the recent samples link): DO NO...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/119526' target='_blank' rel='noopener noreferrer'>Issue #119526: DISABLED test_source_multithreaded_multiple_preexisting_work_in_main_thread_False (__main__.TestProfiler)</a></td><td>needs design decision</td><td>65</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>module: flaky-tests, skipped, oncall: profiler</td><td><div class='wide'>Platforms: dynamo, linux, mac, macos, rocm, slow, asan This test was disabled because it is failing in CI. See recent examples and the most recent trunk workflow logs. Over the past 3 hours, it has been determined flaky in 8 workflow(s) with 24 failures and 8 successes. Debugging instructions (after clicking on the...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189287' target='_blank' rel='noopener noreferrer'>Issue #189287: [vllm] vllm-x-pytorch-test multimodal pooling tests fail with EngineCore &quot;CUDA driver initialization failed&quot; since Triton 3.8 bump (#188251)</a></td><td>high-priority/blocker</td><td>70</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>module: cuda, module: ci, module: tests, triaged, module: regression, oncall: pt2, module: vllm, bot-triaged</td><td><div class='wide'>Job vllm-test / vllm-x-pytorch-test / test (vllm_multi_model_test_28_failure_test, 1, 1, linux.g6.4xlarge.experimental.nvidia.gpu, unstable) Symptom 13 failed / 23 passed / 29 skipped. All 13 failures are vLLM multimodal pooling model tests, and every one dies the same way: vLLM EngineCore fails to start with Runtim...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189282' target='_blank' rel='noopener noreferrer'>Issue #189282: [CPU] operator_benchmark: backward-pass (add, + batchnorm on aarch64) ~200-500x slower since ~Nov 2025 (x86 + aarch64; possible measurement artifact)</a></td><td>needs design decision</td><td>115</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>triage review, module: performance, module: autograd, module: cpu, module: regression, bot-triaged</td><td><div class='wide'>Possible / low-confidence. add_M64_N64_K*_cpu_bwd*_BACKWARD, 9 cases, ~200-400x slower vs the prior baseline, stable across runs. Flagged as low-confidence: a real 400x regression in add backward would be catastrophic and caught in training, so this looks more like a backward-pass measurement artifact than a genuine...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189281' target='_blank' rel='noopener noreferrer'>Issue #189281: [CPU] operator_benchmark: embedding / embeddingbag ~70-130x slower since ~Nov 2025 (x86 + aarch64; possible measurement issue)</a></td><td>needs design decision</td><td>115</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>triage review, module: performance, module: cpu, module: regression, op-bench, module: embedding, bot-triaged</td><td><div class='wide'>Possible regression. embedding and embeddingbag, ~48 cases, ~70-90x slower vs the prior baseline, stable across runs. Flagged as possible because embeddingbag time is roughly flat vs bag count (10 vs 2300 bags both ~8500us), which suggests the number may be dominated by a fixed per-call cost rather than the embeddin...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189283' target='_blank' rel='noopener noreferrer'>Issue #189283: [CPU] operator_benchmark: isolated ~2x regressions (index_add, baddbmm) since ~Nov 2025</a></td><td>high-priority/blocker</td><td>70</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>module: performance, module: cpu, triaged, module: regression, op-bench, bot-triaged</td><td><div class='wide'>Possible / low-confidence. Two isolated single-shape cases flagged by the op-benchmark CPU rebaseline, each stable across two runs but not part of a broader family pattern. index_add__M8_N32_K1_dim0_cpu_dtypetorch.float32 -- ~2.4x baddbmm_B128_M64_N32_K64_cpu_dtypetorch.float32 -- ~2.0x. Note: other baddbmm shapes a...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189280' target='_blank' rel='noopener noreferrer'>Issue #189280: [CPU][quant] operator_benchmark: qcat ~3-4x slower (likely) and fake_quantize ~700x slower (possible) since ~Nov 2025</a></td><td>needs design decision</td><td>65</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>oncall: quantization, bot-triaged</td><td><div class='wide'>Two quantization families flagged by the rebaseline, both stable across runs. qcat (quantized cat), 18 cases, ~3-4x slower -- likely regression. fake_quantize per-tensor/channel (original_kernel/learnable_kernel), ~700x slower -- possible regression or measurement artifact; the magnitude is extreme and needs an owne...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/188833' target='_blank' rel='noopener noreferrer'>Issue #188833: [pipelining] Interleaved1F1B warmup head-of-line bubble from front-loaded RECV</a></td><td>needs design decision</td><td>65</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>oncall: distributed</td><td><div class='wide'>🐛 Describe the bug TL;DR On the _PipelineScheduleRuntime path (which ScheduleInterleaved1F1B now uses), the communication lowering places each RECV_F at the position where its sender is scheduled by a dependency simulation, not at the timestep where the data is actually produced/needed. https://github.com/pytorch/py...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189279' target='_blank' rel='noopener noreferrer'>Issue #189279: [CPU] operator_benchmark: eager transcendental unary (cosh/sinh/atan/log2/log10) ~3-5x slower since ~Nov 2025</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td></td><td><div class='wide'>Likely regression. 8 cases of eager transcendental unary math are ~3-5x slower vs the prior baseline, stable across two runs. Suspected: the vectorized/Sleef CPU math path. Please confirm. sinh_M512_N512_cpu old 81.5 -&gt; 432.0 / 388.9 (~5x) log10__M512_N512_cpu old 11.8 -&gt; 46.2 / 45.1 (~4x) log2__M512_N512_cpu old 11...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189262' target='_blank' rel='noopener noreferrer'>Issue #189262: DISABLED rocm-nightly / linux-noble-rocm-nightly-py3.12-gfx942 / test</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>module: rocm, module: ci, triaged</td><td><div class='wide'>Reason MI300 cluster is having outages, hence capacity is reduced, leading to queueing on 4-GPU MI300 runners. cc @jeffdaily @sunway513 @pruthvistony @ROCmSupport @jataylo @hongxiayang @naromero77amd @pragupta @jerrymannil @xinyazhang @malfet @pytorch/pytorch-dev-infra</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/167822' target='_blank' rel='noopener noreferrer'>Issue #167822: [RFD] Composable Kernel (CK) moving to C++20 — request for feedback/concerns</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>module: build, triaged</td><td><div class='wide'>New Feature for Release Composable Kernel (CK) plans to upgrade its language standard from C++17 → C++20 in the next few months. This will change our default CMake from C++17 to C++20. We’d like to confirm there are no blockers for PyTorch builds and downstreams. Point(s) of contact ThomasNing Release Mode (pytorch/...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/188756' target='_blank' rel='noopener noreferrer'>Issue #188756: [MPS] index_select / F.embedding silently return all-zero rows when the source tensor exceeds 2**31 elements (fixed on main, still broken in release/2.13)</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>triaged, module: 64-bit, module: correctness (silent), module: scatter &amp; gather ops, module: embedding, module: mps, bot-triaged</td><td><div class='wide'>🐛 Describe the bug On MPS, F.embedding / nn.Embedding / torch.index_select silently return all-zero rows (no error, no NaN) for every index i &gt;= ceil(2**31 / dim) when the source weight tensor has more than 2**31 elements. Indices below that boundary return bit-exact rows, so the corruption is partial and easy to mi...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/188836' target='_blank' rel='noopener noreferrer'>Issue #188836: DISABLED test_dropout_parity_dynamic_shapes (__main__.TestDropoutAlignRandomEager)</a></td><td>needs design decision</td><td>65</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>skipped, oncall: pt2</td><td><div class='wide'>This test was disabled because it is failing on main branch (recent examples). cc @chauhang @penguinwu</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/188896' target='_blank' rel='noopener noreferrer'>Issue #188896: DISABLED test_launch_elastic_worker_raise_exception (__main__.ElasticLaunchTest)</a></td><td>needs design decision</td><td>65</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>oncall: distributed, module: flaky-tests, skipped, oncall: r2p</td><td><div class='wide'>Platforms: linux This test was disabled because it is failing in CI. See recent examples and the most recent trunk workflow logs. Over the past 6 hours, it has been determined flaky in 3 workflow(s) with 0 failures and 3 successes. Debugging instructions (after clicking on the recent samples link): DO NOT ASSUME THI...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189267' target='_blank' rel='noopener noreferrer'>Issue #189267: nn.functional.threshold OpInfo reference raises OverflowError on unsigned dtypes with NumPy &gt;= 2.5</a></td><td>has linked PR</td><td>94</td><td><span class='chip-row'><span class='chip'>PR #189273</span></span></td><td>module: nn, module: tests, triaged, module: numpy, module: regression, module: unsigned int, bot-triaged</td><td><div class='wide'>🐛 Describe the bug The nn.functional.threshold OpInfo reference in torch/testing/_internal/common_methods_invocations.py fails for unsigned dtypes under NumPy &gt;= 2.5. test_reference_numerics_{small,normal}_nn_functional_threshold_{cpu,cuda}_uint8 (and the _refs variants) fail with: OverflowError: Python integer -9 o...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189245' target='_blank' rel='noopener noreferrer'>Issue #189245: [cpp_extension] Stale .lock file from FileBaton causes permanent deadlock when compilation process is killed</a></td><td>needs design decision</td><td>65</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>triage review, module: cpp-extensions, module: deadlock, bot-triaged</td><td><div class='wide'>🐛 Describe the bug If the process holding the lock is forcefully terminated (e.g., by OOM Killer, SIGKILL, Ctrl+C, or Slurm job cancellation), the .lock file is never cleaned up. On subsequent runs, all processes detect the stale .lock file and enter an infinite wait loop, causing a permanent deadlock. To Reproduce...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189271' target='_blank' rel='noopener noreferrer'>Issue #189271: [inductor] KeyError in Scheduler.compute_ancestors: self-edge from realizing a reshaped in-place-mutated buffer</a></td><td>has linked PR</td><td>44</td><td><span class='chip-row'><span class='chip'>PR #189288</span></span></td><td>triaged, has workaround, module: custom-operators, module: viewing and reshaping, oncall: pt2, module: dynamic shapes, module: inductor, module: pt2-dispatcher, bot-triaged</td><td><div class='wide'>🐛 Describe the bug torch.compile fails with InductorError: KeyError in Scheduler.compute_ancestors when a custom op mutates freshly-allocated (torch.empty) output buffers in place and those buffers are then reshaped in a way Inductor cannot prove is a free view (under dynamic shapes). This is the shape of a flash_at...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189270' target='_blank' rel='noopener noreferrer'>Issue #189270: [inductor] Emit per-kernel physical op structure (with rematerialization) in provenance</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>triaged, enhancement, oncall: pt2, module: inductor, module: trace, bot-triaged</td><td><div class='wide'>Summary kernel_information.json (#160540, extended in #183952) and the profiler-timeline provenance (#186230, #187328) map each generated kernel to a set of pre/post-grad node names, stack traces, and now extern I/O shapes. What is still missing is the physical, per-kernel op structure: the ordered list of ops that...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/177952' target='_blank' rel='noopener noreferrer'>Issue #177952: Fix onnx tests running on GPU/CUDA</a></td><td>needs reproduction</td><td>32</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>needs reproduction, module: onnx, triaged, OSS contribution wanted, bot-triaged</td><td><div class='wide'>🐛 Describe the bug cd test/onnx python test_pytorch_onnx_onnxruntime_cuda.py -k test_arithmetic_bfp16 onnxruntime.capi.onnxruntime_pybind11_state.NotImplemented: [ONNXRuntimeError] : 9 : NOT_IMPLEMENTED : Could not find an implementation for Add(14) node with name &#x27;/Add&#x27; Versions torch 2.7.1 onnx 1.17.0 onnxruntime...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/188496' target='_blank' rel='noopener noreferrer'>Issue #188496: torch.fft.irfft2 ONNX export produces graph that onnxruntime cannot execute.</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>module: onnx, triaged, has workaround, module: fft, bot-triaged</td><td><div class='wide'>🐛 Describe the bug torch.fft.irfft2 ONNX export now succeeds on nightly (fixed in #164698), but the exported graph cannot be run by onnxruntime in practice. PyTorch&#x27;s dynamo exporter currently produces ONNX IR version 10 graphs. The newest onnxruntime release that supports CUDA 11.8 is onnxruntime-gpu==1.16.3, which...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189261' target='_blank' rel='noopener noreferrer'>Issue #189261: DISABLED periodic-rocm-mi300 / linux-noble-rocm-py3.12-mi300 / test</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>module: rocm, module: ci, triaged, bot-triaged</td><td><div class='wide'>Reason MI300 cluster is having outages, hence capacity is reduced, leading to queueing on 4-GPU MI300 runners. cc @jeffdaily @sunway513 @pruthvistony @ROCmSupport @jataylo @hongxiayang @naromero77amd @pragupta @jerrymannil @xinyazhang @malfet @pytorch/pytorch-dev-infra</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/129131' target='_blank' rel='noopener noreferrer'>Issue #129131: Torch compile initialises CUDA context, even compiling CPU functions</a></td><td>has linked PR</td><td>44</td><td><span class='chip-row'><span class='chip'>PR #186403</span></span></td><td>triaged, oncall: pt2, module: dynamo, module: pt2-dispatcher</td><td><div class='wide'>🐛 Describe the bug When trying to compile a simple function that uses cpu tensors, torch inductor initialises a context on cuda:0. If used in a multiprocessing context (E.G.) a torch dataloader, this quickly results in OOM (simple repro example allocates 500MB of VRAM per process). CUDA_VISIBLE_DEVICES=&quot;&quot; works as a...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189255' target='_blank' rel='noopener noreferrer'>Issue #189255: DISABLED test_delayed_optim_step_offload_true_no_shard_cuda (__main__.TestParityWithDDPCUDA)</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>oncall: distributed, module: rocm, triaged, module: flaky-tests, skipped</td><td><div class='wide'>Platforms: rocm This test was disabled because it is failing in CI. See recent examples and the most recent trunk workflow logs. Over the past 6 hours, it has been determined flaky in 3 workflow(s) with 0 failures and 3 successes. Debugging instructions (after clicking on the recent samples link): DO NOT ASSUME THIN...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/119536' target='_blank' rel='noopener noreferrer'>Issue #119536: DISABLED test_source_multithreaded_complex_work_in_main_thread_True (__main__.TestProfiler)</a></td><td>needs design decision</td><td>65</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>module: flaky-tests, skipped, oncall: profiler</td><td><div class='wide'>Platforms: asan, linux, mac, macos, rocm, slow This test was disabled because it is failing in CI. See recent examples and the most recent trunk workflow logs. Over the past 3 hours, it has been determined flaky in 14 workflow(s) with 42 failures and 14 successes. Debugging instructions (after clicking on the recent...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/179799' target='_blank' rel='noopener noreferrer'>Issue #179799: DISABLED test_grad_has_aux (__main__.FuncTorchHigherOrderOpTestsWithCompiledAutograd)</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>triaged, module: flaky-tests, skipped, oncall: pt2, module: inductor</td><td><div class='wide'>Platforms: linux, slow, rocm This test was disabled because it is failing in CI. See recent examples and the most recent trunk workflow logs. Over the past 6 hours, it has been determined flaky in 6 workflow(s) with 12 failures and 6 successes. Debugging instructions (after clicking on the recent samples link): DO N...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/160740' target='_blank' rel='noopener noreferrer'>Issue #160740: MPS torch.argmax/torch.argmin fail for non-contiguous input</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>triaged, module: correctness (silent), module: mps</td><td><div class='wide'>🕵️‍♂️ Detected with FACTO 🐛 Describe the bug As the title says: import torch def test_argmax(x): device = x.device try: output = torch.argmax(x) print(f&quot; argmax test succeeds for device: {device}. output: {output}&quot;) except Exception as e: print(f&quot; argmax test fails for device: {device}: {e}&quot;) x_cpu_contiguous = torc...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/143954' target='_blank' rel='noopener noreferrer'>Issue #143954: No ONNX function found for &lt;OpOverload(op=&#x27;quantized_decomposed.dequantize_per_channel&#x27;, overload=&#x27;default&#x27;)&gt;</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>module: onnx, triaged, OSS contribution wanted</td><td><div class='wide'>🐛 Describe the bug We tried to leverage per_channel quantization in QAT and exported the trained model in onnx format. model = dummy pytorch model export_model = torch.export.export_for_training( model, example_inputs).module() quantizer = XNNPACKQuantizer().set_global(get_symmetric_quantization_config( is_per_chann...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/187455' target='_blank' rel='noopener noreferrer'>Issue #187455: [MPS] Migrate remaining MPSGraph-backed ops to native Metal to bound graph-cache memory</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>module: performance, module: memory usage, triaged, enhancement, module: mps, bot-triaged</td><td><div class='wide'>Several MPS ops still dispatch through MPSGraph (LookUpOrCreateCachedGraph), which compiles and caches a graph per input shape with no eviction. On shape-varying workloads -- GNNs, variable-length attention, hyperparameter sweeps -- the cache grows without bound and the process eventually OOMs on Apple Silicon. (tor...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/168995' target='_blank' rel='noopener noreferrer'>Issue #168995: MXFP8 + torch.compile error: NotImplementedError: make_reindexer NYI on DtypeView</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>triaged, oncall: pt2, module: inductor</td><td><div class='wide'>🐛 Describe the bug Using the latest pytorch nightly, torch.compile throws an error with torchao mxfp8 MoE training code. Repro Use machine with access to local B200 GPUs Clone https://github.com/pytorch/ao Install from source: USE_CUDA=1 pip install -e . --no-build-isolation Run benchmark script: rm -rf /tmp/torchin...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189244' target='_blank' rel='noopener noreferrer'>Issue #189244: DISABLED test_cat_pointwise (__main__.NumBytesMetricTests)</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>module: rocm, triaged, module: flaky-tests, skipped, oncall: pt2, module: inductor</td><td><div class='wide'>Platforms: rocm This test was disabled because it is failing in CI. See recent examples and the most recent trunk workflow logs. Over the past 6 hours, it has been determined flaky in 4 workflow(s) with 4 failures and 4 successes. Debugging instructions (after clicking on the recent samples link): DO NOT ASSUME THIN...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189223' target='_blank' rel='noopener noreferrer'>Issue #189223: DISABLED test_vmap_free_tensor (__main__.FuncTorchHigherOrderOpTestsWithCompiledAutograd)</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>triaged, module: flaky-tests, skipped, oncall: pt2, module: inductor</td><td><div class='wide'>Platforms: linux, slow This test was disabled because it is failing in CI. See recent examples and the most recent trunk workflow logs. Over the past 6 hours, it has been determined flaky in 4 workflow(s) with 8 failures and 4 successes. Debugging instructions (after clicking on the recent samples link): DO NOT ASSU...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/119490' target='_blank' rel='noopener noreferrer'>Issue #119490: DISABLED test_source_multithreaded_complex_work_in_main_thread_False (__main__.TestProfiler)</a></td><td>needs design decision</td><td>65</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>module: flaky-tests, skipped, oncall: profiler</td><td><div class='wide'>Platforms: dynamo, asan, linux, mac, macos, rocm, slow This test was disabled because it is failing in CI. See recent examples and the most recent trunk workflow logs. Over the past 3 hours, it has been determined flaky in 4 workflow(s) with 12 failures and 4 successes. Debugging instructions (after clicking on the...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/187289' target='_blank' rel='noopener noreferrer'>Issue #187289: DISABLED test_col2im_uint32_overflow (__main__.TestMPS)</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>triaged, module: flaky-tests, module: macos, skipped, module: mps</td><td><div class='wide'>Platforms: mac, macos This test was disabled because it is failing in CI. See recent examples and the most recent trunk workflow logs. Over the past 6 hours, it has been determined flaky in 3 workflow(s) with 4 failures and 3 successes. Debugging instructions (after clicking on the recent samples link): DO NOT ASSUM...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/152632' target='_blank' rel='noopener noreferrer'>Issue #152632: DISABLED test_torchvision_models_efficientnet_v2_l (__main__.TestVisionTracing)</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>triaged, module: flaky-tests, skipped, module: fx</td><td><div class='wide'>Platforms: asan, linux, slow This test was disabled because it is failing in CI. See recent examples and the most recent trunk workflow logs. Over the past 3 hours, it has been determined flaky in 3 workflow(s) with 3 failures and 3 successes. Debugging instructions (after clicking on the recent samples link): DO NO...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/179879' target='_blank' rel='noopener noreferrer'>Issue #179879: DISABLED test_vmap_vmap_cpu (__main__.TestComposabilityCPU)</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>triaged, module: flaky-tests, skipped, oncall: pt2, module: functorch</td><td><div class='wide'>Platforms: dynamo This test was disabled because it is failing in CI. See recent examples and the most recent trunk workflow logs. Over the past 6 hours, it has been determined flaky in 12 workflow(s) with 12 failures and 12 successes. Debugging instructions (after clicking on the recent samples link): DO NOT ASSUME...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189239' target='_blank' rel='noopener noreferrer'>Issue #189239: [Inductor]Unable to apply layout optimization on convolution backward causes training performance regression.</a></td><td>needs design decision</td><td>115</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>triage review, module: performance, module: convolution, module: regression, module: memory format, oncall: pt2, module: inductor, bot-triaged</td><td><div class='wide'>🐛 Describe the bug decide_layout_opt() only counts aten.convolution.default nodes when deciding whether to enable channels-last layout optimization. When AOTAutograd splits a model into separate forward and backward graphs, the backward graph may contain only convolution_backward without a corresponding forward conv...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189236' target='_blank' rel='noopener noreferrer'>Issue #189236: [Feature Request] Add &#x27;symmetric&#x27; padding_mode (repeats boundary sample)</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>module: nn, triaged, function request, module: padding, bot-triaged</td><td><div class='wide'>🚀 The feature, motivation and pitch PyTorch&#x27;s current &#x27;reflect&#x27; padding_mode uses half-sample reflection, i.e. it does not repeat the boundary sample: input: ABCDEF reflect: DCB|ABCDEF|EDC (boundaries not repeated) I&#x27;d like to add a &#x27;symmetric&#x27; padding_mode that does repeat the boundary sample: input: ABCDEF symmetr...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/185405' target='_blank' rel='noopener noreferrer'>Issue #185405: [Inductor] AssertionError: len(index) == len(stride) in ir.py when using torch.compile(mode=&quot;max-autotune&quot;) on CPU BMM/GEMM</a></td><td>high-priority/blocker</td><td>70</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>module: crash, module: cpu, triaged, module: complex, oncall: pt2, module: inductor, bot-triaged</td><td><div class='wide'>🐛 Describe the bug When compiling an attention-like function containing complex numbers (torch.polar, torch.view_as_complex) and batched matrix multiplication (torch.matmul) on CPU with torch.compile(mode=&quot;max-autotune&quot;), the compiler crashes with an AssertionError. The crash occurs during the C++ backend&#x27;s BMM/GEMM...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/187175' target='_blank' rel='noopener noreferrer'>Issue #187175: [feature request] Autotune TMA and non-TMA path for pointwise and reduction kernels</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>module: performance, feature, module: cuda, triaged, oncall: pt2, module: inductor, bot-triaged</td><td><div class='wide'>🚀 The feature, motivation and pitch Summary When autotuning is enabled for GEMM kernels, both the TMA (Tensor Memory Accelerator) path and the non-TMA path are included as candidates and benchmarked against each other. This ensures the fastest path is always selected based on actual measured performance. However, fo...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189228' target='_blank' rel='noopener noreferrer'>Issue #189228: [symm_mem] one/two-shot all-reduce deadlocks when issued from multiple streams: ops lack channel plumbing</a></td><td>needs design decision</td><td>65</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>oncall: distributed, bot-triaged, oncall: distributed infra</td><td><div class='wide'>🐛 Describe the bug symm_mem::two_shot_all_reduce_ (and the one-shot / multimem variants) synchronize the group with barriers that require the device execution order of the ops to match their group issue order. CUDA does not order kernels across streams, so issuing these ops from more than one stream — combined with...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/147326' target='_blank' rel='noopener noreferrer'>Issue #147326: torch.export.export fails when one input is a class inheriting from torch.nn.Module</a></td><td>needs design decision</td><td>89</td><td><span class='chip-row'><span class='chip'>PR #185804</span></span></td><td>oncall: pt2, export-triaged, oncall: export</td><td><div class='wide'>🐛 Describe the bug transformers.cache_utils.DynamicCache inherits from torch.nn.Module. It seems to confuse torch.export.export and gives the following error: File &quot;.../site-packages/torch/export/_trace.py&quot;, line 1697, in _export_to_aten_ir_make_fx raise UserError(UserErrorType.CONSTRAINT_VIOLATION, str(e)) # noqa:...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189144' target='_blank' rel='noopener noreferrer'>Issue #189144: [XPU][B580] `flex-attn-causal` performance drop with `last_level_cache_size` cache clear between runs</a></td><td>needs design decision</td><td>115</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>triage review, module: performance, module: regression, has workaround, oncall: pt2, module: inductor, module: xpu, module: flex attention, bot-triaged</td><td><div class='wide'>This PR #184567 introduced regression in a few cases for flex-attn-causal with triton, for example: H_q=16, H_kv=2, N_CTX_q=1, N_CTX_kv=1024, D_HEAD_qk=128, D_HEAD_v=128 case dropped from 0.81 -&gt; 0.41 TFLOPS More details: intel/intel-xpu-backend-for-triton#7302 TL;DR; 18MB of cache clear between runs seems like not...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/187811' target='_blank' rel='noopener noreferrer'>Issue #187811: DISABLED test_combo_kernel_dynamic_scale_rblock (__main__.ComboKernelTests)</a></td><td>has linked PR</td><td>44</td><td><span class='chip-row'><span class='chip'>PR #189224</span></span></td><td>triaged, skipped, module: xpu</td><td><div class='wide'>Platforms: xpu This test was disabled because it is failing on main branch (recent examples). cc @gujinghui @EikanWang @fengyuan14 @guangyey cc @gujinghui @EikanWang @fengyuan14 @guangyey</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/187810' target='_blank' rel='noopener noreferrer'>Issue #187810: DISABLED test_combo_kernel_dynamic_scale_rblock (__main__.ComboKernelTestsPerSubkernelBlocks)</a></td><td>has linked PR</td><td>44</td><td><span class='chip-row'><span class='chip'>PR #189224</span></span></td><td>triaged, skipped, module: xpu</td><td><div class='wide'>Platforms: xpu This test was disabled because it is failing on main branch (recent examples). cc @gujinghui @EikanWang @fengyuan14 @guangyey</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/163650' target='_blank' rel='noopener noreferrer'>Issue #163650: [DO NOT CLOSE] Autorevert actions shadow mode stream</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>triaged</td><td><div class='wide'>This PR can be subscribed by anyone who is interested. Autorevert bot should post here, as comments, all its Autorevert decisions while running in shadow mode. This is, anytime Autorevert decided to act to revert a PR (this means faster than a human). The goal of this PR is to enable the team to quickly troubleshoot...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189221' target='_blank' rel='noopener noreferrer'>Issue #189221: DISABLED test_nested_dict_partial_load (__main__.TestCheckpointer)</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>module: rocm, triaged, module: flaky-tests, skipped, oncall: distributed checkpointing</td><td><div class='wide'>Platforms: rocm This test was disabled because it is failing in CI. See recent examples and the most recent trunk workflow logs. Over the past 6 hours, it has been determined flaky in 3 workflow(s) with 0 failures and 3 successes. Debugging instructions (after clicking on the recent samples link): DO NOT ASSUME THIN...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/119514' target='_blank' rel='noopener noreferrer'>Issue #119514: DISABLED test_source_multithreaded_close_in_scope_work_in_main_thread_True (__main__.TestProfiler)</a></td><td>needs design decision</td><td>65</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>module: flaky-tests, skipped, oncall: profiler</td><td><div class='wide'>Platforms: asan, linux, mac, macos, rocm, slow This test was disabled because it is failing in CI. See recent examples and the most recent trunk workflow logs. Over the past 3 hours, it has been determined flaky in 26 workflow(s) with 78 failures and 26 successes. Debugging instructions (after clicking on the recent...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/179742' target='_blank' rel='noopener noreferrer'>Issue #179742: DISABLED test_grad_freevar_python_scalar (__main__.FuncTorchHigherOrderOpTestsWithCompiledAutograd)</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>triaged, module: flaky-tests, skipped, oncall: pt2, module: inductor</td><td><div class='wide'>Platforms: linux, slow, rocm This test was disabled because it is failing in CI. See recent examples and the most recent trunk workflow logs. Over the past 6 hours, it has been determined flaky in 9 workflow(s) with 18 failures and 9 successes. Debugging instructions (after clicking on the recent samples link): DO N...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/180213' target='_blank' rel='noopener noreferrer'>Issue #180213: DISABLED test_vmap_vjp_cpu (__main__.TestComposabilityCPU)</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>triaged, module: flaky-tests, skipped, oncall: pt2, module: functorch</td><td><div class='wide'>Platforms: dynamo This test was disabled because it is failing in CI. See recent examples and the most recent trunk workflow logs. Over the past 6 hours, it has been determined flaky in 7 workflow(s) with 7 failures and 7 successes. Debugging instructions (after clicking on the recent samples link): DO NOT ASSUME TH...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/76185' target='_blank' rel='noopener noreferrer'>Issue #76185: Ensure custom Function are correct in double backward setting</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>module: autograd, triaged, enhancement, actionable</td><td><div class='wide'>Problem statement Today, if intermediary Tensors are saved during the forward and used during the backward, the double backward will be silently wrong because their contribution will be ignored. A small repro is: import torch from torch.autograd import Function, gradcheck, gradgradcheck class MyFn(Function): @static...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/187073' target='_blank' rel='noopener noreferrer'>Issue #187073: torch.compile recompiles when input alternates between AsyncCollectiveTensor and Tensor</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>module: performance, oncall: distributed, triaged, has workaround, module: fsdp, oncall: pt2, module: dynamo, module: guards, bot-triaged, oncall: distributed parallelisms</td><td><div class='wide'>torch.compile recompiles when input alternates between AsyncCollectiveTensor and Tensor Summary When a compiled region&#x27;s input alternates between AsyncCollectiveTensor (ACT) and a resolved Tensor across steps, Dynamo guards on the tensor class and recompiles each time the class changes. With FSDP + TP this happens d...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189189' target='_blank' rel='noopener noreferrer'>Issue #189189: Subtraction between two `torch.complex64` tensors fails on modules compiled with `aoti_compile_and_package`</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>triaged, module: complex, has workaround, oncall: pt2, oncall: export, module: aotinductor, bot-triaged</td><td><div class='wide'>🐛 Describe the bug I am trying to export an nn.Module that involves manipulation of complex-valued tensors with torch.export.export and then compiling with torch._inductor.aoti_compile_and_package. Compilation succeeds but I run into errors when loading the package and running. In the following minimal reproducing e...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/186796' target='_blank' rel='noopener noreferrer'>Issue #186796: weakrefs don&#x27;t when tensor swap is used</a></td><td>has linked PR</td><td>44</td><td><span class='chip-row'><span class='chip'>PR #186801</span></span></td><td>triaged, oncall: pt2, module: dynamo, module: guards, bot-triaged</td><td><div class='wide'>🐛 Describe the bug I thought #180566 would fix it but I was wrong. A live torch.compile&#x27;d module keeps weakrefs on its parameters (Dynamo guards). swap_tensors() refuses to operate on tensors that have weakrefs. The follow should reproduce the issue: &quot;&quot;&quot;Minimal reproducer for why group offloading cannot use torch.ut...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/50282' target='_blank' rel='noopener noreferrer'>Issue #50282: Add MultiScheduler</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>module: optimizer, triaged, enhancement</td><td><div class='wide'>🚀 Feature I would like to have a MultiScheduler that allows me to use multiple schedulers after each other, with the same optimizer. Motivation It is not uncommon to have a learning rate schedule that looks like the following example: First have a burn in learning rate for 1000-2000 batches (eg. quadratic or exponen...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/167725' target='_blank' rel='noopener noreferrer'>Issue #167725: `torch._inductor.utils.fresh_cache` fails due to persistant process pool</a></td><td>has linked PR</td><td>44</td><td><span class='chip-row'><span class='chip'>PR #184007</span></span></td><td>triaged, oncall: pt2, module: inductor, repro:agent:h100:repro_fail:not_reproduced, repro:agent:h100:incorrect</td><td><div class='wide'>🐛 Describe the bug Consider this code: import torch from torch._inductor.utils import fresh_cache def my_addmm(x, a, b, alpha, beta): return torch.addmm(x, a, b, alpha=beta, beta=alpha) x = torch.randn((128, 128)).cuda().half() a = torch.randn(128, 128).cuda().half() b = torch.randn(128, 128).cuda().half().t() with...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/187781' target='_blank' rel='noopener noreferrer'>Issue #187781: [RFC] Topology-aware nn.Module primitives for memory optimization in torch.compile</a></td><td>has linked PR</td><td>62</td><td><span class='chip-row'><span class='chip'>PR #187790</span></span></td><td>feature, module: nn, module: memory usage, triaged, needs research, oncall: pt2, module: compile ux, bot-triaged</td><td><div class='wide'>Hi PyTorch team, I am working on a research prototype called TopologyAwareJointOptimizer. I would like feedback on whether topology-aware memory optimization is a direction that should be explored around torch.compile, torch.nn, torchao, or an out-of-tree PyTorch backend. The problem I am trying to address is traini...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189161' target='_blank' rel='noopener noreferrer'>Issue #189161: [Inductor] Remove Redundant NaN Propagation in `max` op in Softmax</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>module: performance, triaged, enhancement, oncall: pt2, module: inductor, bot-triaged</td><td><div class='wide'>🚀 The feature, motivation and pitch Redundant NaN Propagation in max op in non-onlined Softmax Issue Current Inductor softmax can do redundant NaN propagation in the softmax-internal max reduction. For the fallback/persistent softmax form, the computation is effectively: xmax = max(x, dim=-1, keepdim=True) xexp = ex...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189157' target='_blank' rel='noopener noreferrer'>Issue #189157: GHA stats artifact downloader does not skip mismatched workflow run attempts</a></td><td>has linked PR</td><td>44</td><td><span class='chip-row'><span class='chip'>PR #189158</span></span></td><td>module: ci, triaged, bot-triaged</td><td><div class='wide'>🐛 Describe the bug tools.stats.upload_stats_lib._download_artifact() is intended to skip GitHub-hosted fallback artifacts whose embedded runattemptN token does not match the requested workflow_run_attempt. Today the mismatch branch only prints a Skipping ... message. After that, the function continues to download th...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189151' target='_blank' rel='noopener noreferrer'>Issue #189151: DISABLED test_pending_fusion_pro_and_epi_use_async_compile_False (__main__.TestPrologueFusion)</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>module: rocm, triaged, module: flaky-tests, skipped, oncall: pt2, module: inductor</td><td><div class='wide'>Platforms: rocm This test was disabled because it is failing in CI. See recent examples and the most recent trunk workflow logs. Over the past 6 hours, it has been determined flaky in 4 workflow(s) with 4 failures and 4 successes. Debugging instructions (after clicking on the recent samples link): DO NOT ASSUME THIN...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189133' target='_blank' rel='noopener noreferrer'>Issue #189133: DISABLED test_vmap (__main__.FuncTorchHigherOrderOpTestsWithCompiledAutograd)</a></td><td>has linked PR</td><td>44</td><td><span class='chip-row'><span class='chip'>PR #189286</span></span></td><td>triaged, module: flaky-tests, skipped, oncall: pt2, module: inductor</td><td><div class='wide'>Platforms: linux, rocm, slow This test was disabled because it is failing in CI. See recent examples and the most recent trunk workflow logs. Over the past 6 hours, it has been determined flaky in 12 workflow(s) with 24 failures and 12 successes. Debugging instructions (after clicking on the recent samples link): DO...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/152944' target='_blank' rel='noopener noreferrer'>Issue #152944: DISABLED test_compiler_collectives_automatic_dynamic_tensor (__main__.TestMultiProc)</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>triaged, module: flaky-tests, skipped, oncall: pt2, module: dynamic shapes, module: dynamo, PT2-Bug-Bash</td><td><div class='wide'>Platforms: inductor, rocm This test was disabled because it is failing in CI. See recent examples and the most recent trunk workflow logs. Over the past 3 hours, it has been determined flaky in 3 workflow(s) with 3 failures and 3 successes. Debugging instructions (after clicking on the recent samples link): DO NOT A...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/119364' target='_blank' rel='noopener noreferrer'>Issue #119364: DISABLED test_source_multithreaded_close_in_scope_work_in_main_thread_False (__main__.TestProfiler)</a></td><td>needs design decision</td><td>65</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>module: flaky-tests, skipped, oncall: profiler</td><td><div class='wide'>Platforms: asan, dynamo, linux, mac, macos, rocm, slow This test was disabled because it is failing in CI. See recent examples and the most recent trunk workflow logs. Over the past 3 hours, it has been determined flaky in 10 workflow(s) with 30 failures and 10 successes. Debugging instructions (after clicking on th...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/186865' target='_blank' rel='noopener noreferrer'>Issue #186865: DISABLED test_dynamic_shape_backend_eager_device_cpu (__main__.TestPackage)</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>triaged, module: flaky-tests, skipped, oncall: pt2, module: dynamo</td><td><div class='wide'>Platforms: linux This test was disabled because it is failing in CI. See recent examples and the most recent trunk workflow logs. Over the past 6 hours, it has been determined flaky in 3 workflow(s) with 3 failures and 3 successes. Debugging instructions (after clicking on the recent samples link): DO NOT ASSUME THI...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/180195' target='_blank' rel='noopener noreferrer'>Issue #180195: DISABLED test_vjp_vmap_cpu (__main__.TestComposabilityCPU)</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>triaged, module: flaky-tests, skipped, oncall: pt2, module: functorch</td><td><div class='wide'>Platforms: dynamo This test was disabled because it is failing in CI. See recent examples and the most recent trunk workflow logs. Over the past 6 hours, it has been determined flaky in 3 workflow(s) with 3 failures and 3 successes. Debugging instructions (after clicking on the recent samples link): DO NOT ASSUME TH...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189137' target='_blank' rel='noopener noreferrer'>Issue #189137: [Sub-RFC] PrivateUse1 Inductor codegen registration interfaces (RFC 3/3)</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>feature, triaged, module: backend, oncall: pt2, module: inductor, oncall: export, module: aotinductor, module: PrivateUse1, bot-triaged</td><td><div class='wide'>TL;DR This sub-RFC addresses the third slice of the umbrella RFC: providing registration interfaces for Inductor codegen so that a PrivateUse1 backend can be auto-discovered, configured, compiled, and run without monkeypatching upstream Inductor code. Four things are addressed: (1) backend discovery via entry-point...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189136' target='_blank' rel='noopener noreferrer'>Issue #189136: [Sub-RFC] PrivateUse1 Dynamo tracing via registry splicing (RFC 2/3)</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>triaged, enhancement, oncall: pt2, module: dynamo, module: PrivateUse1, bot-triaged</td><td><div class='wide'>TL;DR This sub-RFC addresses the second slice of the umbrella RFC: making Dynamo&#x27;s six hardcoded lookup tables recognize PrivateUse1 device types and runtime functions by splicing them from the device interface registry. We introduce get_interface_for_device__PrivateUse1() — an iterable that returns only PrivateUse1...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189135' target='_blank' rel='noopener noreferrer'>Issue #189135: [Sub-RFC] PrivateUse1 device identity/capability in torch.compile (RFC 1/3)</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>feature, triaged, module: backend, oncall: pt2, module: inductor, module: dynamo, module: PrivateUse1, bot-triaged</td><td><div class='wide'>TL;DR This sub-RFC addresses the first slice of the umbrella RFC: making torch.compile query the DeviceInterface contract for device identity (is_gpu) and capabilities (is_triton_capable, get_multi_processor_count, exposes_streams) instead of hardcoded cuda/xpu/mps lists. We introduce register_interface_for_device_P...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189138' target='_blank' rel='noopener noreferrer'>Issue #189138: [RFC] DeviceInterface: Registry Support for PrivateUse1 Backends in torch.compile</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>feature, triaged, large, module: backend, oncall: pt2, module: inductor, module: dynamo, module: PrivateUse1, bot-triaged</td><td><div class='wide'>TL;DR This RFC document aims to propose and discuss making PrivateUse1 (out-of-tree) backends first-class citizens in torch.compile. Our focus is on eliminating the hardcoded cuda/xpu/mps device lists and if-branches that permeate the Dynamo and Inductor codebase, replacing them with a unified registry-based contrac...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/75171' target='_blank' rel='noopener noreferrer'>Issue #75171: torch.jit.load fails when path contains non-ascii characters</a></td><td>needs design decision</td><td>65</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>oncall: jit, module: windows</td><td><div class='wide'>🐛 Describe the bug openai/CLIP#227 torch.jit.load fails when path contains non-ascii characters import torch torch.jit.load(&#x27;C:\\Users\\流星暴雨/.cache/clip\\ViT-B-16.pt&#x27;, &quot;cpu&quot;) Traceback (most recent call last): File &quot;&lt;string&gt;&quot;, line 1, in &lt;module&gt; File &quot;[C:\Users\]()流星暴雨\AppData\Local\Programs\Python\Python39\lib\sit...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/188892' target='_blank' rel='noopener noreferrer'>Issue #188892: nvidia-cudnn-cu13==9.20.0.48 bundled by torch is incomplete → CUDNN_STATUS_SUBLIBRARY_VERSION_MISMATCH</a></td><td>has linked PR</td><td>106</td><td><span class='chip-row'><span class='chip'>PR #189043</span></span></td><td>needs reproduction, module: binaries, module: crash, module: cudnn, module: convolution, triaged, has workaround, bot-triaged</td><td><div class='wide'>🐛 Describe the bug torch==2.12.1+cu130 pins nvidia-cudnn-cu13==9.20.0.48, whose libcudnn_graph.so.9 dlopens libcudnn_engines_tensor_ir.so by bare soname - but that wheel doesn&#x27;t ship it (it&#x27;s only in ≥9.23). With no other cuDNN around this fails silently and falls back. But if a different system cuDNN 9.x is on the...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/186867' target='_blank' rel='noopener noreferrer'>Issue #186867: DISABLED test_group_norm_backward_large_input_float16_trigger_32bit_overflow_True_mps_float16 (__main__.TestConsistencyMPS)</a></td><td>has linked PR</td><td>44</td><td><span class='chip-row'><span class='chip'>PR #189215</span></span></td><td>triaged, module: flaky-tests, module: macos, skipped, module: mps</td><td><div class='wide'>Platforms: mac, macos This test was disabled because it is failing in CI. See recent examples and the most recent trunk workflow logs. Over the past 6 hours, it has been determined flaky in 19 workflow(s) with 38 failures and 19 successes. Debugging instructions (after clicking on the recent samples link): DO NOT AS...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/188798' target='_blank' rel='noopener noreferrer'>Issue #188798: DISABLED test_linearize_jvp_fn (__main__.FuncTorchHigherOrderOpTestsWithCompiledAutograd)</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>triaged, module: flaky-tests, skipped, oncall: pt2, module: inductor</td><td><div class='wide'>Platforms: linux, rocm, slow This test was disabled because it is failing in CI. See recent examples and the most recent trunk workflow logs. Over the past 6 hours, it has been determined flaky in 3 workflow(s) with 6 failures and 3 successes. Debugging instructions (after clicking on the recent samples link): DO NO...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/119357' target='_blank' rel='noopener noreferrer'>Issue #119357: DISABLED test_source_multithreaded_basic_work_in_main_thread_True (__main__.TestProfiler)</a></td><td>needs design decision</td><td>65</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>module: flaky-tests, skipped, oncall: profiler</td><td><div class='wide'>Platforms: linux, rocm, asan, mac, macos, slow This test was disabled because it is failing in CI. See recent examples and the most recent trunk workflow logs. Over the past 3 hours, it has been determined flaky in 11 workflow(s) with 33 failures and 11 successes. Debugging instructions (after clicking on the recent...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/176966' target='_blank' rel='noopener noreferrer'>Issue #176966: DISABLED test_trace_return_namedtuple (__main__.TestFX)</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>triaged, module: flaky-tests, skipped, module: fx, oncall: pt2</td><td><div class='wide'>Platforms: dynamo This test was disabled because it is failing in CI. See recent examples and the most recent trunk workflow logs. Over the past 6 hours, it has been determined flaky in 9 workflow(s) with 9 failures and 9 successes. Debugging instructions (after clicking on the recent samples link): DO NOT ASSUME TH...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/124152' target='_blank' rel='noopener noreferrer'>Issue #124152: DISABLED test_torchvision_models_efficientnet_v2_m (__main__.TestVisionTracing)</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>triaged, module: flaky-tests, skipped, module: fx</td><td><div class='wide'>Platforms: asan, linux, rocm, slow This test was disabled because it is failing in CI. See recent examples and the most recent trunk workflow logs. Over the past 3 hours, it has been determined flaky in 2 workflow(s) with 2 failures and 2 successes. Debugging instructions (after clicking on the recent samples link):...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/188332' target='_blank' rel='noopener noreferrer'>Issue #188332: Inductor: allgather fusion creates rank-specific cache that crashes when loaded by other ranks</a></td><td>high-priority/blocker</td><td>70</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>high priority, triaged, oncall: pt2</td><td><div class='wide'>🐛 Describe the bug Minimal repro: Inductor FX graph cache reuses rank-specific code. Run twice with two GPUs. The first run cold-compiles and should pass; the second run reloads the shared warm FX graph cache and should fail because one rank can load the other rank&#x27;s device/rank-specific generated wrapper. rm -rf /t...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/144965' target='_blank' rel='noopener noreferrer'>Issue #144965: RuntimeError &quot;global alloc not supported yet&quot; when using TorchScript optimization.</a></td><td>needs design decision</td><td>115</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>high priority, triage review, oncall: jit</td><td><div class='wide'>🐛 Describe the bug Calling the forward method on TorchScript models can, under some specific conditions, raise a RuntimeError with message: &quot;Global alloc not supported yet.&quot;. I think this is linked to an old issue: #69078, however, I managed to consistently reproduce this error. The code that reproduces the bug is q...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/188987' target='_blank' rel='noopener noreferrer'>Issue #188987: DISABLED test_optimizer_non_static_param (__main__.LoggingTests)</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>module: logging, module: tests, triaged, skipped, oncall: pt2, module: xpu, bot-triaged</td><td><div class='wide'>Platforms: xpu This test was disabled because it is failing on main branch (recent examples). cc @gujinghui @EikanWang @fengyuan14 @guangyey cc @mruberry @chauhang @penguinwu @gujinghui @EikanWang @fengyuan14 @guangyey</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/188970' target='_blank' rel='noopener noreferrer'>Issue #188970: [MPS] SIGSEGV in MPSStream::copy with triton.cudagraphs (reduce-overhead/max-autotune) on deep fp16 graphs</a></td><td>has linked PR</td><td>106</td><td><span class='chip-row'><span class='chip'>PR #181720</span></span></td><td>needs reproduction, module: crash, triaged, module: macos, module: mps, module: inductor, bot-triaged</td><td><div class='wide'>🐛 Describe the bug For a family of deep fp16 inference models, torch.compile on MPS deterministically crashes the process (SIGSEGV, exit 139) whenever the compile mode enables triton.cudagraphs, and works (bit-exact correct outputs) whenever it does not: configuration triton.cudagraphs result eager — ✅ OK torch.comp...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189126' target='_blank' rel='noopener noreferrer'>Issue #189126: Report a bug, as requested by the error message</a></td><td>needs design decision</td><td>115</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>triage review, module: crash, module: cuda, module: memory usage, bot-triaged</td><td><div class='wide'>🐛 Describe the bug terminate called after throwing an instance of &#x27;c10::Error&#x27; what(): inserted INTERNAL ASSERT FAILED at &quot;/pytorch/c10/cuda/CUDACachingAllocator.cpp&quot;:3252, ... Versions curl -sL https://raw.githubusercontent.com/pytorch/pytorch/main/torch/utils/collect_env.py | python command returns: Collecting env...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/187093' target='_blank' rel='noopener noreferrer'>Issue #187093: [inductor] baddbmm with broadcast bias: autotune never considers the bmm + pointwise-bias decomposition (1.3-1.45x left on the table at MoE expert-GEMM shapes)</a></td><td>has linked PR</td><td>44</td><td><span class='chip-row'><span class='chip'>PR #187106</span><span class='chip'>PR #189127</span></span></td><td>module: performance, triaged, enhancement, oncall: pt2, module: inductor, bot-triaged</td><td><div class='wide'>🐛 Describe the bug For baddbmm with a broadcast bias (bias[B, 1, N], stride [N, 0, 1] after broadcasting), Inductor&#x27;s max-autotune chooses between (a) Triton templates and (b) the extern ATen kernel — but never considers the decomposition bmm(x, w) + bias, which is faster than both of its existing choices whenever t...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189121' target='_blank' rel='noopener noreferrer'>Issue #189121: [inductor] Single-config reductions extended by _dynamic_scale_rblock are permanently uncacheable — re-benchmark {R0_BLOCK, R0_BLOCK/2} every process; near-tied winners flip perf and numerics</a></td><td>needs design decision</td><td>89</td><td><span class='chip-row'><span class='chip'>PR #189124</span></span></td><td>triage review, module: performance, module: determinism, module: reductions, oncall: pt2, module: inductor, module: compile-time, bot-triaged</td><td><div class='wide'>🐛 Describe the bug Follow-up to #181789 (related: #172819 / #173391): one more case where _dynamic_scale_rblock mutates the autotune candidate set after the cache contract is frozen, defeating autotune_local_cache=True. Verified on current nightly. check_autotune_cache only creates an AutotuneCache when there is mor...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/185839' target='_blank' rel='noopener noreferrer'>Issue #185839: torch.linalg.qr (float64) repeated in a loop crashes the process (SIGSEGV, exit 139) on ROCm 7.x</a></td><td>high-priority/blocker</td><td>70</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>module: crash, module: rocm, triaged, module: regression, has workaround, module: linear algebra, module: CUDACachingAllocator, bot-triaged</td><td><div class='wide'>🐛 Describe the bug Calling torch.linalg.qr on a float64 matrix repeatedly in a loop causes the process to be killed by SIGSEGV (exit code 139) after ~31 iterations on ROCm 7.x (AMD Instinct MI300A). No Python exception is raised — the process dies silently with no traceback. The crash is specific to torch.linalg.qr...</div></td></tr><tr><td><a href='https://github.com/pytorch/pytorch/issues/189046' target='_blank' rel='noopener noreferrer'>Issue #189046: Treat PrivateUse1 as CUDA-like under env override to reuse CUDA optimization paths in shared `at::native` operators</a></td><td>needs maintainer decision</td><td>28</td><td><span class='chip-row'><span class='muted'>none</span></span></td><td>feature, module: autograd, triaged, module: advanced indexing, module: dispatch, module: backend, module: reductions, module: PrivateUse1, bot-triaged</td><td><div class='wide'>🚀 The feature, motivation and pitch Allow CUDA-like PrivateUse1 backends to opt into existing CUDA optimization paths in shared at::native operators via an environment variable, avoiding code duplication and performance loss. 🚀 Feature Both this issue and #176715 share the same goal: third-party privateuse1 backends...</div></td></tr></tbody></table></div></section></div>