ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-03-07 04:20:03 +00:00
Code Issues Packages Projects Releases Wiki Activity

WIP - Qwen3-MoE (and hopefully all others) working

Browse Source 
But when I say here and in the previous commit "working",
I mean PP is working. TG is still broken.
This commit is contained in:
Kawrakow
2026-01-12 11:55:01 +02:00
parent 4e4fabf0b4
commit 3a848fc48c
1 changed files with 1 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
ggml/src/ggml-cuda.cu
View File

@@ -2689,13 +2689,9 @@ static int ggml_cuda_moe_up_gate_unary(ggml_backend_cuda_context & ctx, ggml_ten
ggml_swiglu_oai_cuda_f32((const float *)dst_up_gate_contiguous.get() + dst->ne[0], (const float *)dst_up_gate_contiguous.get(),
(float *)dst->data, ggml_nelements(dst), dst->ne[0], src0_1->ne[1], src0_1->ne[1],
1.702f, 7.0f, stream);
//ggml_swiglu_oai_cuda_f32((const float *)dst_up_gate_contiguous.get(), (const float *)dst_up_gate_contiguous.get() + dst->ne[0],
// (float *)dst->data, ggml_nelements(dst), dst->ne[0], src0_1->ne[0], src0_1->ne[0],
// 1.702f, 7.0f, stream);
} else {
ggml_fused_mul_unary(ctx, (ggml_unary_op)dst->op_params[0], ggml_nelements(dst), dst->ne[0],
(const float *)dst_up_gate_contiguous.get(),
(float *)dst->data);
(const float *)dst_up_gate_contiguous.get(), (float *)dst->data);
}
}
CUDA_CHECK(cudaGetLastError());