We read every piece of feedback, and take your input very seriously.
To see all available qualifiers, see our documentation.
1 parent a91e779 commit 9f60666Copy full SHA for 9f60666
1 file changed
ggml/src/ggml-openvino/ggml-openvino.cpp
@@ -932,6 +932,15 @@ static bool is_op_unsupported_case(const ggml_tensor * op) {
932
}
933
break;
934
935
+ case GGML_OP_DIV: {
936
+ // The GPU plugin can fuse broadcast DIV into the preceding FFN GEMM path
937
+ // and produce infs for per-channel scale vectors. Keep those DIVs on CPU
938
+ // until the fused GPU kernel is reliable. (falied case llama-arch-test mpt)
939
+ if (op->src[1]->ne[0] == 1 && op->src[1]->ne[1] == 1 && op->src[1]->ne[2] == 1 && op->src[1]->ne[3] == 384) {
940
+ return true;
941
+ }
942
+ break;
943
944
case GGML_OP_SUM_ROWS: {
945
// if the input is PERMUTE skip
946
if (op->src[0]->op == GGML_OP_PERMUTE) {
0 commit comments