mirror of
https://github.com/ikawrakow/ik_llama.cpp.git
synced 2026-02-24 07:04:11 +00:00
Also this one
This commit is contained in:
@@ -508,7 +508,7 @@ void cuda_bailingmoev2_experts(ggml_backend_cuda_context & ctx, ggml_tensor * ds
|
||||
|
||||
{
|
||||
const dim3 block_dims(WARP_SIZE, 1, 1);
|
||||
const dim3 block_nums(1, nrows, 1);
|
||||
const dim3 block_nums(nrows, 1, 1);
|
||||
k_apply_mask<<<block_nums, block_dims, 0, ctx.stream()>>>((float *)topk_src->data, discarded_groups.get(), n_discarded_groups, n_per_group, ne00);
|
||||
CUDA_CHECK(cudaGetLastError());
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user