Int4 Qkv Scale Concat#
- cudaError_t trt_edgellm::kernel::launchGptqInt4QkvScaleConcat(
- void const *qScalesGN,
- int32_t qWidth,
- void const *kScalesGN,
- int32_t kWidth,
- void const *vScalesGN,
- int32_t vWidth,
- int32_t numGroups,
- void *output,
- cudaStream_t stream
Concatenate Q/K/V GPTQ scales into
[G,Nq+Nk+Nv].