]> git.djapps.eu Git - pkg/ggml/sources/whisper.cpp/commit
sycl: support reordered Q4_K/Q5_K/Q6_K MoE MUL_MAT_ID (llama/24452)
authorFrosty40 <redacted>
Tue, 16 Jun 2026 05:35:00 +0000 (00:35 -0500)
committerGeorgi Gerganov <redacted>
Fri, 19 Jun 2026 09:53:43 +0000 (12:53 +0300)
commitfa204c82f4f644871052f938cd8a5e89cdeea1dd
treedd7f0472c0bb10d6b4e24abf96bdea86a377743c
parent79f88a1104ef95aad161af3d6b912ccb794bccc4
sycl: support reordered Q4_K/Q5_K/Q6_K MoE MUL_MAT_ID (llama/24452)

* sycl: support reordered Q4_K and Q5_K MoE MUL_MAT_ID

Extend reordered-weight handling to fused MoE MUL_MAT_ID for Q4_K and Q5_K expert tensors and add Q5_K reordered DMMV coverage. Unsupported 3D reorder cases now fall back instead of aborting.

* sycl: extend MoE reorder to Q6_K mul_mat_id
ggml/src/ggml-sycl/dmmv.cpp
ggml/src/ggml-sycl/ggml-sycl.cpp
ggml/src/ggml-sycl/mmvq.cpp
ggml/src/ggml-sycl/mmvq.hpp