All issues

Search results

[Windows] CMake preset llm-debug-vulkan selects MSVC instead of Clang, causing XNNPACK compilation failures

module: vulkanIssues related to the Vulkan delegate and code under backends/vulkan/Issues related to the Vulkan delegate and code under backends/vulkan/

Status: Open.

#22084 In pytorch/executorch;

· mtavenrath opened on Aug 24, 2026

Remove the need for symlinks when building ExecuTorch on Windows

Status: Open.

#22081 In pytorch/executorch;

· mtavenrath opened on Aug 24, 2026

to_edge_transform_and_lower does not constant-fold parameter-only subgraphs, so weight_norm'd convolutions never reach the delegate (941x on one module)

Status: Open.

#22078 In pytorch/executorch;

· john-rocky opened on Aug 24, 2026

macOS wheel: libquantized_ops_aot_lib.dylib never loads, so PT2E models fail later with "Missing out variants"

Status: Open.

#22071 In pytorch/executorch;

· john-rocky opened on Aug 24, 2026

PT2E quantization silently does nothing when the model's forward is under torch.no_grad()

Status: Open.

#22069 In pytorch/executorch;

· john-rocky opened on Aug 24, 2026

_convert_scalars_to_attrs writes every lifted scalar as float32, retyping integer subgraphs and breaking prepare_pt2e

Status: Open.

#22062 In pytorch/executorch;

· john-rocky opened on Aug 23, 2026

QNN-delegated static-cache decoder produces correct output via native C++ Module, but degenerate/repeating output via Android JNI Module (same .pte, same device)

module: qnnIssues related to Qualcomm's QNN delegate and code under backends/qualcomm/Issues related to Qualcomm's QNN delegate and code under backends/qualcomm/

Status: Open.

Bug

#22061 In pytorch/executorch;

· aramvyan opened on Aug 23, 2026

qwen3_5 example config leaves use_sdpa_with_kv_cache off, which costs 2x decode

Status: Open.

#22045 In pytorch/executorch;

· john-rocky opened on Aug 22, 2026

8da4w silently skips layers that do not divide the group size, and 8da8w is implemented but rejected by config validation

Status: Open.

#22044 In pytorch/executorch;

· john-rocky opened on Aug 22, 2026

no_rope_layer_interval is declared but never applied, and SmolLM3's own config sets it

Status: Open.

#22043 In pytorch/executorch;

· john-rocky opened on Aug 22, 2026

MLX backend: missing aten.native_group_norm and upsample_nearest2d fragment a diffusion UNet into 28 subgraphs

module: mlxIssues related to MLX Backend: Metal-accelerated inference on Apple SiliconIssues related to MLX Backend: Metal-accelerated inference on Apple Silicon

Status: Open.

#22017 In pytorch/executorch;

· msluszniak opened on Aug 21, 2026

MLX iOS: whisper-small bf16 uses ~6.7x its file size in RSS and is jetsam-killed, while every sibling stays under 2x

module: mlxIssues related to MLX Backend: Metal-accelerated inference on Apple SiliconIssues related to MLX Backend: Metal-accelerated inference on Apple Silicon

Status: Open.

#22016 In pytorch/executorch;

· msluszniak opened on Aug 21, 2026