All issues

Search results

VulkanQuantizer dynamic activation quantization is per tensor, and very lossy on transformer encoders

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.

#22431 In pytorch/executorch;

· msluszniak opened on Sep 2, 2026

Vulkan weight-only int8 linear dispatches to shaders that do not exist (linear_qcs8w_*)

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.

#22429 In pytorch/executorch;

· msluszniak opened on Sep 2, 2026

MLX backend returns wrong numbers for causal attention when query and key lengths differ

Status: Open.

#22426 In pytorch/executorch;

· shoumikhin opened on Sep 2, 2026

MethodMeta::uses_backend() crashes (null pointer deref) on non-delegated programs

Status: Open.

#22404 In pytorch/executorch;

· bharqav opened on Sep 1, 2026

Vulkan _convert_scalars_to_attrs lifts integer scalars as float32, breaking integer indexing

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.

#22365 In pytorch/executorch;

· AxelNoun opened on Aug 31, 2026

Core ML llama export ignores enable_state and preserve_sdpa

Status: Open.

#22352 In pytorch/executorch;

· shoumikhin opened on Aug 31, 2026

[ET-VK] softmax_buffer normalises each row with 4 threads and aggregates serially in every thread

Status: Open.

#22351 In pytorch/executorch;

· msluszniak opened on Aug 31, 2026

[ET-VK] Reductions dispatch a fixed 16 threads regardless of the reduction length

Status: Open.

#22350 In pytorch/executorch;

· msluszniak opened on Aug 31, 2026

Feedback about Quick Start Pathway

Status: Open.

#22335 In pytorch/executorch;

· rosesshipka1505-svg opened on Aug 30, 2026

Building a backend tailored to Espressif systems

Status: Open.

#22334 In pytorch/executorch;

· CodeKitty-max opened on Aug 30, 2026

[ET-VK] Embedding with vocab > 16384 silently returns wrong values on the texture path; force_fp16 breaks all-MiniLM-L6-v2

Status: Open.

#22333 In pytorch/executorch;

· msluszniak opened on Aug 30, 2026

[ET-VK] Binary ops with a 0-dimensional output abort with std::out_of_range

Status: Open.

#22331 In pytorch/executorch;

· msluszniak opened on Aug 30, 2026