ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-02-24 07:04:11 +00:00
Code Issues Packages Projects Releases Wiki Activity

Disable graphs without -fmoe

Browse Source
This commit is contained in:
Iwan Kawrakow
2025-08-15 08:05:22 +03:00
parent 8a83e1f083
commit 54084f28c0
1 changed files with 3 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
ggml/src/ggml-cuda.cu
View File

@@ -3272,10 +3272,11 @@ static bool check_node_graph_compatibility_and_refresh_copy_ops(ggml_backend_cud
#endif
}
if (node->op == GGML_OP_MUL_MAT_ID && node->ne[2] != 1) {
if (node->op == GGML_OP_MUL_MAT_ID && (node->ne[2] != 1 || node->src[2]->ne[0] != 1)) {
use_cuda_graph = false; // This node type is not supported by CUDA graph capture
#ifndef NDEBUG
GGML_CUDA_LOG_DEBUG("%s: disabling CUDA graphs due to unsupported node type\n", __func__);
GGML_CUDA_LOG_DEBUG("%s(%s): disabling CUDA graphs due to unsupported node type %ld %ld\n",
__func__, node->src[0]->name, node->ne[2], node->src[2]->ne[0]);
#endif
}
if (node->op == GGML_OP_MOE_FUSED_UP_GATE) {