Repository navigation

Issues

New issue

Label tracking meta-issue (edit me to get automatically CC'ed on issues! cc bot)
#7679 · mergennachin opened on Jan 15, 2025
[v1.5.0] Release Schedule and Tracker
#22188 · JacobSzwejbka opened on Aug 26, 2026
5

Feedback about Tools
pytorch/executorch#23706 · opened on Oct 12, 2026
1
Portable floor_divide returns non-whole numbers for floats and wrong values for a zero divisor
Bug
pytorch/executorch#23684 · opened on Oct 11, 2026
Portable remainder kernel: wrong sign for integers and float64 rounded to float
Bug
pytorch/executorch#23682 · opened on Oct 11, 2026
torch.var_mean and torch.std_mean fail in to_edge: aten.var_mean.correction is not core ATen and has no core decomposition
Bug
pytorch/executorch#23680 · opened on Oct 10, 2026
Prompts of max_seq_len tokens fail to prefill: export_llm bounds the KV-cache token input at max_seq_len - 1 but publishes get_max_seq_len = max_seq_len
Bug
pytorch/executorch#23678 · opened on Oct 10, 2026
[v1.6.0] Release Schedule and Tracker
pytorch/executorch#23608 · opened on Oct 8, 2026
QNN sharded LLM export fails: ResolveDebugHandle is not the last edge pass (dep_table entry overwritten by SplitGraph registration)
Bug
pytorch/executorch#23580 · opened on Oct 8, 2026
Python bindings: Module path gives device-tagged unplanned outputs host memory
pytorch/executorch#23553 · opened on Oct 7, 2026
UNSTABLE trunk / test-samsung-models-linux / linux-job
pytorch/executorch#23507 · opened on Oct 6, 2026
1
Samsung: test_yolo26 fails to lower, AI LiteCore cannot export a CUCustomOp layer
pytorch/executorch#23481 · opened on Oct 6, 2026
5
yolo26 example breaks on portable with ultralytics newer than 8.4.14
pytorch/executorch#23453 · opened on Oct 6, 2026
2
Add Arm export and tests for MLPerf Tiny streaming wakeword
Feature
pytorch/executorch#23440 · opened on Oct 5, 2026
Add Cortex-M export and tests for MLPerf Tiny streaming wakeword
Feature
pytorch/executorch#23434 · opened on Oct 5, 2026
Arm backend: int16 (A16W8) rsqrt lowered to a 513-point TOSA TABLE is several times off for small inputs, which breaks A16W8 RMSNorm/LayerNorm on Ethos-U85
Task
pytorch/executorch#23356 · opened on Oct 2, 2026
3
XNNPACK never delegates slice_copy with stride != 1, but op-support.csv only excludes zero-dim/dynamic shapes
pytorch/executorch#23258 · opened on Sep 29, 2026
[RFC] ExecuTorch Persisting Device Specialized Delegate Artifacts
pytorch/executorch#23192 · opened on Sep 28, 2026
6
[QNN] Enable ConvTranspose + BatchNorm fusion after #23170
pytorch/executorch#23185 · opened on Sep 28, 2026
1
[Tracking] Complete audio task coverage across CPU, GPU, and NPU backends
pytorch/executorch#23164 · opened on Sep 25, 2026
7
Qualcomm: 8-bit per-channel weight scales are floored at the 16-bit eps, and the HTP miscomputes near-zero channels
Task
pytorch/executorch#23160 · opened on Sep 25, 2026
1
[cpu kernels] native_layer_norm: layer_norm_scalar returns NaN on large-mean rows; Half/BF16 at N>=256 slow after #23153
pytorch/executorch#23159 · opened on Sep 25, 2026
2
Qualcomm: 16a16w quantization is much less accurate than 16a8w
Bug
pytorch/executorch#23108 · opened on Sep 24, 2026
4
Enable native Kev inference with the WebGPU backend
pytorch/executorch#23099 · opened on Sep 23, 2026
2
Enable native Kev inference with the Vulkan backend
pytorch/executorch#23098 · opened on Sep 23, 2026
2
Enable native Kev inference with the AOTI CUDA backend
pytorch/executorch#23097 · opened on Sep 23, 2026
2
Qualcomm: HTP hangs in fp16 on a dilated conv when the dilated extent exceeds the input and cin is large
Bug
pytorch/executorch#23096 · opened on Sep 23, 2026
3

Navigation Menu

Repository navigation

Issues

Issues

Feedback about Tools

Portable floor_divide returns non-whole numbers for floats and wrong values for a zero divisor

Portable remainder kernel: wrong sign for integers and float64 rounded to float

torch.var_mean and torch.std_mean fail in to_edge: aten.var_mean.correction is not core ATen and has no core decomposition

Prompts of max_seq_len tokens fail to prefill: export_llm bounds the KV-cache token input at max_seq_len - 1 but publishes get_max_seq_len = max_seq_len

[v1.6.0] Release Schedule and Tracker

QNN sharded LLM export fails: ResolveDebugHandle is not the last edge pass (dep_table entry overwritten by SplitGraph registration)

Python bindings: Module path gives device-tagged unplanned outputs host memory

UNSTABLE trunk / test-samsung-models-linux / linux-job

Samsung: test_yolo26 fails to lower, AI LiteCore cannot export a CUCustomOp layer

yolo26 example breaks on portable with ultralytics newer than 8.4.14

Add Arm export and tests for MLPerf Tiny streaming wakeword

Add Cortex-M export and tests for MLPerf Tiny streaming wakeword

Arm backend: int16 (A16W8) `rsqrt` lowered to a 513-point TOSA TABLE is several times off for small inputs, which breaks A16W8 RMSNorm/LayerNorm on Ethos-U85

XNNPACK never delegates slice_copy with stride != 1, but op-support.csv only excludes zero-dim/dynamic shapes

[RFC] ExecuTorch Persisting Device Specialized Delegate Artifacts

[QNN] Enable ConvTranspose + BatchNorm fusion after #23170

[Tracking] Complete audio task coverage across CPU, GPU, and NPU backends

Qualcomm: 8-bit per-channel weight scales are floored at the 16-bit eps, and the HTP miscomputes near-zero channels

[cpu kernels] native_layer_norm: layer_norm_scalar returns NaN on large-mean rows; Half/BF16 at N>=256 slow after #23153

Qualcomm: 16a16w quantization is much less accurate than 16a8w

Enable native Kev inference with the WebGPU backend

Enable native Kev inference with the Vulkan backend

Enable native Kev inference with the AOTI CUDA backend

Qualcomm: HTP hangs in fp16 on a dilated conv when the dilated extent exceeds the input and cin is large