happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Change the input and ouput node shape of MUL_MAT operator

This commit is contained in:
zhanmyz 2025-03-06 01:49:14 +08:00 committed by Mustafa Cavus
parent f37fa21a5c
commit 246a2d1021
1 changed files with 14 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
ggml/src/ggml-openvino.cpp
View File

@ -531,14 +531,25 @@ void ggml_backend_openvino_mul_mat(struct ggml_tensor * dst) {
auto B = reshape_src1;
auto batched_matmul = std::make_shared<ov::op::v0::MatMul>(B, A, false, false);
auto model = std::make_shared<ov::Model>(ov::NodeVector{ batched_matmul },
std::vector<int64_t> final_output_shape = {static_cast<int64_t>(dst->ne[2]),
static_cast<int64_t>(dst->ne[1]),
static_cast<int64_t>(dst->ne[0])};
auto reshape_output = std::make_shared<ov::op::v1::Reshape>(
batched_matmul,
ov::op::v0::Constant::create(ov::element::i64, {3}, final_output_shape),
false
);
auto model = std::make_shared<ov::Model>(ov::NodeVector{ reshape_output },
ov::ParameterVector{ param_src0, param_src1 });
ov::Tensor tensor_src0{ ov::element::f16, orig_shape_src0, src0->data };
ov::Tensor tensor_src1{ ov::element::f32, orig_shape_src1, src1->data };
ov::Shape output_shape = { static_cast<size_t>(dst->ne[0]),
ov::Shape output_shape = { static_cast<size_t>(dst->ne[2]),
static_cast<size_t>(dst->ne[1]),
static_cast<size_t>(dst->ne[2]) };
static_cast<size_t>(dst->ne[0]) };
ov::Tensor tensor_dst(ov::element::f32, output_shape, dst->data);
ov::Core core;