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

Always apply Eliminate_ZP to fix GPU compile issue on some platforms

This commit is contained in:
Yu, Zijun 2025-09-28 22:21:23 +08:00 committed by Mustafa Cavus
parent fdadca1e89
commit 973a80fd02
2 changed files with 4 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
ggml/src/ggml-openvino/openvino/pass/eliminate_zp.cpp
View File

@ -80,6 +80,7 @@ EliminateZeroPoints::EliminateZeroPoints() {
std::shared_ptr<ov::op::v0::Constant> new_constant;
// TODO improve performance
if (data_type == ov::element::u4) {
auto data_values = data_constant->cast_vector<uint8_t>();
std::vector<int8_t> adjusted_values(total_elements);

6
ggml/src/ggml-openvino/openvino/translate_session.cpp
View File

@ -233,9 +233,9 @@ std::shared_ptr<Model> TranslateSession::apply_transformations(std::shared_ptr<M
manager.register_pass<ov::pass::MakeStateful>(kv_param_res_pairs);
}
if (ggml_model_decoder->is_static()) {
manager.register_pass<pass::EliminateZeroPoints>();
}
// if (ggml_model_decoder->is_static()) {
manager.register_pass<pass::EliminateZeroPoints>();
// }
manager.run_passes(model);
}
return model;