Gptq Activation Permutation

cudaError_t trt_edgellm::kernel::launchGptqActivationPermutation(

int32_t const *groupIndex,

int32_t *permutation,

int32_t K,

int32_t numGroups,

int32_t groupSize,

cudaStream_t stream

)

Build the stable GPTQ activation permutation that groups source channels by g_idx while preserving source-channel order within each group.