openvino
[GPU] Enable fused MoE for trinity-mini AFMoE (sub-128 group _size + routed_scaling_factor)
#35684
Merged
Go
Login via GitHub
Home
Pricing
FAQ
Install
Login
via GitHub
Overview
Commits
8
Changes
View On
GitHub
[GPU] Enable fused MoE for trinity-mini AFMoE (sub-128 group _size + routed_scaling_factor)
#35684
e-ddykim
merged 8 commits into
openvinotoolkit:master
from
andrew-k-park:moe-3gemm-trinity-mini-afmoe
andrew-k-park
requested a review
2 days ago
andrew-k-park
requested a review
2 days ago
andrew-k-park
added this to the
2026.2
milestone
2 days ago
github-actions
added
category: GPU
andrew-k-park
force pushed
from
478be6ab
to
0e803248
2 days ago
andrew-k-park
force pushed
from
0e803248
to
e23c6500
2 days ago
mg-intel
assigned
v-Golubev
2 days ago
mg-intel
assigned
EgorDuplensky
2 days ago
v-Golubev
commented on 2026-05-06
andrew-k-park
force pushed
from
e23c6500
to
7a4ceee8
2 days ago
andrew-k-park
requested a review
from
v-Golubev
2 days ago
v-Golubev
approved these changes on 2026-05-06
andrew-k-park
assigned
e-ddykim
2 days ago
andrew-k-park
force pushed
from
b634ee37
to
dcb63b60
2 days ago
e-ddykim
commented on 2026-05-07
geunhwan
added
Code Freeze
e-ddykim
commented on 2026-05-07
andrew-k-park
force pushed
from
800caf03
to
e6de0f6e
2 days ago
andrew-k-park
force pushed
from
e6de0f6e
to
4e62e0c4
2 days ago
e-ddykim
approved these changes on 2026-05-07
e-ddykim
enabled auto-merge
2 days ago
ahnyoung-paul
approved these changes on 2026-05-07
andrew-k-park
force pushed
from
4e62e0c4
to
78315375
1 day ago
andrew-k-park
force pushed
from
596f53e9
to
c0bf111a
1 day ago
[GPU] fuse_moe_3gemm_compressed: support routed_scaling_factor in sig…
e1ac87a5
[GPU] moe_3gemm_swiglu_mlp: generalize FAKE_GROUP_SIZE for sub-128 gr…
1aa51c82
[GPU] moe_3gemm tests: add group_size=64 cases and adjust tolerance
55d7a151
[GPU] fuse_moe_3gemm_compressed: address review comments
3a5ff122
[GPU] fuse_moe_3gemm_compressed_test: fix gcc maybe-uninitialized war…
56888da2
[GPU] moe_3gemm_gpu_test: limit sub-128 group_size cases to SIGMOID_BIAS
b8fabec4
[GPU] moe_3gemm_gpu_test: drop unnecessary 3x tolerance for sub-128 g…
cf18af3b
[GPU] MoE 3GEMM SwiGLU: add ELEMS_PER_LANE == 2 inner-loop variant
c0bf111a
e-ddykim
merged
af349e14
into master
1 day ago
Login to write a write a comment.
Login via GitHub
Reviewers
ahnyoung-paul
e-ddykim
v-Golubev
Assignees
EgorDuplensky
v-Golubev
e-ddykim
Labels
category: GPU
Code Freeze
Milestone
2026.2
Login to write a write a comment.
Login via GitHub