feat(KERNEL-QUANT-CIQ-GEMM-ROCM-IQUANT): port IQ4_XS/IQ3_XXS to ROCm - #3029

Open

joral wants to merge 12 commits into

mudler:mainfrom

joral:row/KERNEL-QUANT-CIQ-GEMM-ROCM-IQUANT-w1

Commits on Sep 12, 2026

feat(KERNEL-QUANT-CIQ-GEMM-ROCM-IQUANT): port IQ4_XS/IQ3_XXS to ROCm

joral
authored and
mudler
committed
eeecccf
docs(KERNEL-QUANT-CIQ-GEMM-ROCM-IQUANT): name IQ4_XS/IQ3_XXS in the ROCm keep-quant line

joral
authored and
mudler
committed
6bc041c
spec(KERNEL-QUANT-CIQ-GEMM-ROCM-IQUANT): define contribution integration
mudler
committed
213160b
test(KERNEL-QUANT-CIQ-GEMM-ROCM-IQUANT): cover IQ3_XXS loader admission
mudler
committed
1e77579
docs(KERNEL-QUANT-CIQ-GEMM-ROCM-IQUANT): correct footprint units
mudler
committed
c44147b
spec(KERNEL-QUANT-CIQ-GEMM-ROCM-IQUANT): pin the repaired gate baseline
mudler
committed
73f5627
spec(KERNEL-QUANT-CIQ-GEMM-ROCM-IQUANT): seal device tables
mudler
committed
c29fd6c
test(KERNEL-QUANT-CIQ-GEMM-ROCM-IQUANT): compare device codebooks
mudler
committed
92e4e50
test(KERNEL-QUANT-CIQ-GEMM-ROCM-IQUANT): snapshot HIP tables
mudler
committed
19af6b5
docs(KERNEL-QUANT-CIQ-GEMM-ROCM-IQUANT): record the ROCm IQ4_XS and IQ3_XXS coverage

joral
authored and
mudler
committed
173b396
docs(KERNEL-QUANT-CIQ-GEMM-ROCM-IQUANT): record the operator's own gate reproduction

joral
authored and
mudler
committed
0c8ffb9
fix(KERNEL-QUANT-CIQ-GEMM-ROCM-IQUANT): close the fused-MoE TEST_CASE before the IQ4_XS block
mudler
committed
3b5a7e7