llama.cpp
sycl: add grouped MoE XMX GEMM
#29245
Merged

sycl: add grouped MoE XMX GEMM #29245

cwriter
cwriter cwriter requested a review 32 days ago
github-actions github-actions added ggml
github-actions github-actions added SYCL
Titaniumtown
arthw
arthw commented on 2026-09-23
cwriter
cwriter cwriter force pushed 30 days ago
github-actions github-actions added documentation
cwriter
sycl: add a grouped MoE XMX GEMM, gathering IQ weights into the tiles
950f1c4a
cwriter cwriter force pushed to 950f1c4a 29 days ago
cwriter Merge branch 'master' into grouped_moe_xmx
53095e0b
arthw
arthw commented on 2026-09-29
arthw
arthw commented on 2026-09-29
sycl: pick the XMX dequant-GEMM joint_matrix shape per device
53e19649
sycl: address review of the XMX dequant-GEMM docs and naming
eea286fa
sycl: type-aware combination selection for the XMX dequant-GEMM
995a7ff7
cwriter
arthw
arthw
cwriter
sycl: add GGML_SYCL_DYNAMIC_PRECISION for the XMX dequant-GEMM
9e8a1a4b
sycl: decode q8_0 and the k-quants in the grouped XMX dequant-GEMM
9a7d2829
sycl: reorder k-quant MoE weights before the grouped XMX GEMM
11f866cc
cwriter
cwriter Merge branch 'master' into grouped_moe_xmx
0ba8f0f8
arthw
arthw
arthw approved these changes on 2026-10-02
arthw arthw added merge ready
ggerganov
Titaniumtown
Titaniumtown
Titaniumtown commented on 2026-10-07
address feedback
f5b30785
Titaniumtown
Titaniumtown approved these changes on 2026-10-08
Titaniumtown Titaniumtown merged 847f447c into master 15 days ago
Titaniumtown

Login to write a write a comment.

Login via GitHub

Reviewers
Assignees
No one assigned
Labels
Milestone