COMMITS
August 4, 2026
R
Arm backend: add VGF runtime accuracy comparisons (#21396)
Rob Elliott committed
S
August 3, 2026
J
Clean up remaining TARGETS files (#21462)
Jon Janzen committed
J
(Re-re-re-land) Update XNNPACK to github revision f6486e3e1d"" (#21399)
Jacob Stevens committed
J
migrate runtime/test (with PACKAGE gating) (#21433)
Jon Janzen committed
A
Fix broken torchao documentation links (#21547)
Anthony Shoumikhin committed
V
NXP backend: added support for scalar args of elementary ops (#21011)
Vaclav Novak committed
M
XNNPACK: Run recommended pre-partition transforms in AOT flow (#20446)
Måns Nilsson committed
D
Qualcomm AI Engine Direct - Preserve partition metadata on single-output delegate getitem nodes (#20805)
DannyYuyang-quic committed
W
Qualcomm AI Engine Direct - Pytest Refactor for Passes (#21003)
winskuo-quic committed
August 2, 2026
B
Arm backend: Add serialised xlarge VKML model suite (#21518)
Baris committed
August 1, 2026
P
P
[ExecuTorch][llm] Add MoE quantized custom op documentation (#21539)
pytorchbot committed
P
[ExecuTorch][llm] Fuse w1+w3 into single GEMM in quantized_moe_ffn (#21538)
pytorchbot committed
P
[ExecuTorch][llm] Add numerical kernel tests for quantized_moe_ffn (#21537)
pytorchbot committed
P
[ExecuTorch][llm] Add source-transform and Meta-kernel tests for MoE op (#21536)
pytorchbot committed
P
[ExecuTorch][llm] Add MoE source transformation with INT4 quantized packing (#21535)
pytorchbot committed
P
P
[ExecuTorch][llm] Add quantized_moe_ffn custom op with C++ kernel (#21533)
pytorchbot committed
A
backends/mlx: runtime MoE expert-sort for decode (issue #20554) (#20685)
Axel.Cffrd.Dnty committed
July 31, 2026
R
Bump torchao to 03ca489dc and nightly wheel to dev20260729 (#21476)
RJ Ascani committed
K
[MLX] Grow the KV-cache pool on demand (#21525)
Kiymet Akdemir committed
S
[ET-VK][ops] Add buffer unfold support
Stephen Jia committed
S
[ET-VK][ops] Add log10 support
Stephen Jia committed
S
[ET-VK][ops] Add scalar comparison operators
Stephen Jia committed
K
[MLX] Add off-graph KV-cache flat runtime (#21501)
Kiymet Akdemir committed
J
migrate backends/vulkan/tools/gpuinfo (TARGETS+BUCK with similar rules) (#21432)
Jon Janzen committed
J
migrate backends/vulkan/test (fbcode python_unittest + xplat fb_native) (#21431)
Jon Janzen committed
J
drop empty TARGETS in backends/mediatek/runtime (#21430)
Jon Janzen committed
S