Fp16 Moe Assemble

cudaError_t trt_edgellm::kernel::launchFp16MoeSourceBatch(

uint8_t const *const *firstSources,

uint8_t const *const *secondSources,

int32_t count,

void *output,

int32_t sourceRows,

int32_t sourceColumns,

int32_t outputRows,

int32_t outputColumns,

Fp16MoeSourceType sourceType,

cudaStream_t stream

)

Cast and stack checkpoint matrices directly into an FP16 MoE buffer. When secondSources is non-null, each source pair is emitted in the Fp16MoePlugin’s fixed 64-row up/gate interleave.