Repository Issue Activity (beta)

#22062 `_convert_scalars_to_attrs` writes every lifted scalar as float32, retyping integer subgraphs and breaking `prepare_pt2e`

#22061 QNN-delegated static-cache decoder produces correct output via native C++ Module, but degenerate/repeating output via Android JNI Module (same .pte, same device)

#22043 no_rope_layer_interval is declared but never applied, and SmolLM3's own config sets it

#22045 qwen3_5 example config leaves use_sdpa_with_kv_cache off, which costs 2x decode

#22044 8da4w silently skips layers that do not divide the group size, and 8da8w is implemented but rejected by config validation

#21855 `export_llm` Core ML builds fail at execute: "must be an MLState" even though `take_over_mutable_buffer=False`

#21827 [Muse Glimmer][CUDA] SM80+PTX vision kernel exceeds RTX 5090 shared-memory limit; native SM120 export OOMs

#21805 iOS swiftpm-1.4.0 prebuilts: HF tokenizers with lookahead regex abort the app (signal 6) — regex_lookahead still missing

#21804 Deterministic SIGSEGV in Method::execute_instruction via EventTracerProfileOpScope destructor — QNN restore-fails→DLC-fallback path

#21904 AtenToCortexMPass crashes with "unsupported param type, call_function" when aten.linear.default is in preserve_ops

#22017 MLX backend: missing aten.native_group_norm and upsample_nearest2d fragment a diffusion UNet into 28 subgraphs

#22016 MLX iOS: whisper-small bf16 uses ~6.7x its file size in RSS and is jetsam-killed, while every sibling stays under 2x

#20486 ARM Ethos-U: Operators not properly quantized when `cat` is present

#22013 [CoreML] Build failure on Ubuntu 26.04 LTS (ExecuTorch 1.4.1, GCC 15.2.0): missing <cstdint> in memory_buffer.hpp

#22012 SDPA decomposition does not preserve intermediate dtype for fp16/bf16 inputs

#20268 URML (open robot intent language): a validated-action gate above on-device policy inference (request for comment)

#20701 [RFC] dflash

#21880 Docs for Qualcomm scripts use outdated (removed) CLI flags

#21870 QNN DeepLabV3 example opens too many files

#21943 aten.bmm, aten.mul, aten._softmax doesn't get quantized/replaced with Cortex-M kernels in Transformer model

#21586 QNN HTP fails to initialize on Snapdragon 8 Elite (SM8750P) — native “No Snapdragon SOC detected” / transport error 4000

#21829 XNNPACK: quantized slice fails to build when its output scale differs from its input

#21951 CoreML delegate emits unloadable multi-method program (0x23 InvalidProgram); MLX and XNNPACK load the same graphs fine

#21950 aten._fft_r2c has no portable kernel: program builds but fails to load with 0x14 OperatorMissing

#19045 Arm backend bug: `TOSAPartitioner.partition()` makes export order depend on Python hash randomization