ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-03-07 20:40:02 +00:00
Code Issues Packages Projects Releases Wiki Activity

WIP

Browse Source
This commit is contained in:
Iwan Kawrakow
2025-11-02 10:00:26 +02:00
parent 332c4d6680
commit 60d56fa2d0
1 changed files with 2 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
ggml/src/ggml-cuda.cu
View File

@@ -3097,7 +3097,7 @@ static bool ggml_cuda_compute_forward(ggml_backend_cuda_context & ctx, struct gg
ggml_are_same_shape(dst, cgraph->nodes[i+1]->src[1]) &&
cgraph->nodes[i+1] == cgraph->nodes[i+2]->src[0] &&
ops_are_same_device(cgraph, i, i+2)) {
//printf("Fusing add->add->fused_rms of %s, %s, %s\n", dst->name, cgraph->nodes[i+1]->name, cgraph->nodes[i+2]->name);
printf("Fusing add->add->fused_rms of %s, %s, %s\n", dst->name, cgraph->nodes[i+1]->name, cgraph->nodes[i+2]->name);
ggml_cuda_op_fused_add_add_rms_norm(ctx, dst, cgraph->nodes[i+1], cgraph->nodes[i+2]);
i += 2;
}
@@ -3249,6 +3249,7 @@ static bool ggml_cuda_compute_forward(ggml_backend_cuda_context & ctx, struct gg
// printf("=== Fused rms_norm(%s)\n", dst->name);
// for (int j = 1; j <= 6; ++j) printf(" %s(%s)\n", ggml_op_name(cgraph->nodes[i+j]->op), cgraph->nodes[i+j]->name);
//}
// Disabled because currently there is something wrong in the fused kernel implementation
if (false && ENABLE_FUSION && i + 4 < cgraph->n_nodes &&
cgraph->nodes[i+1]->op == GGML_OP_VIEW &&
cgraph->nodes[i+2]->op == GGML_OP_FUSED_RMS_NORM &&