2026.5

New issue

Open

No due date

•Last updated Nov 6, 2026

96% complete

[GPU] CM u2 weight-compressed FullyConnected (XMX/DPAS)
Unread
#38894 opened Oct 15, 2026 · Updated Nov 5, 2026
19
[CPU] Add GroupedMatmul initial implementation
Unread
#37444 opened Aug 14, 2026 · Updated Oct 22, 2026Draft
10
[CPU] Fix PagedAttention FP16 BRGEMM buffer packing
Unread
#38886 opened Oct 15, 2026 · Updated Oct 28, 2026
6
[CPU] Keep large Q2 expert weights compressed via exact u8 widening
Unread
#39029 opened Oct 20, 2026 · Updated Oct 23, 2026Draft
6
[GPU]enable property to select kernel
Unread
#37880 opened Sep 3, 2026 · Updated Nov 3, 2026
17
[GPU] Fix weightless cache reload for MoE offload
Unread
#39024 opened Oct 20, 2026 · Updated Nov 6, 2026
48
[GPU] Fix MoE offload slot-id buffer freed while kernels still read it
Unread
#38545 opened Oct 1, 2026 · Updated Oct 30, 2026
14
[GPU] Sample ze
Unread
#39323 opened Oct 29, 2026 · Updated Nov 6, 2026
15
[GPU] Int8 activations for u2 FullyConnected on Xe2/Xe3
Unread
#39320 opened Oct 29, 2026 · Updated Nov 5, 2026Draft
3
[GPU] support scale/zp as input
Unread
#38204 opened Sep 16, 2026 · Updated Nov 7, 2026
52
Gpu fc weights layout
Unread
#39011 opened Oct 20, 2026 · Updated Nov 6, 2026
13
Cm u2 fc gnk native
Unread
#39044 opened Oct 20, 2026 · Updated Nov 4, 2026Draft
5
[NPUW] Bound-check deserialized attention mask_idx before port-vector subscript
Unread
#38785 opened Oct 12, 2026 · Updated Nov 6, 2026
14
[NPUW] Support reset state for Whisper
Unread
#39421 opened Nov 2, 2026 · Updated Nov 6, 2026
9
[ci-doctor] Retry JS test model downloads on transient network timeouts
Unread
#39435 opened Nov 2, 2026 · Updated Nov 3, 2026
1
[NPUW] Support attention sink in HFA and Pyramid
Unread
#37647 opened Aug 26, 2026 · Updated Nov 6, 2026
105
[CPU] Fix model input overwritten by subsequent output binding
Unread
#39523 opened Nov 5, 2026 · Updated Nov 6, 2026
2

Issues