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

Fix Q4_1

This commit is contained in:
Yu, Zijun 2025-08-08 11:07:10 +08:00 committed by Mustafa Cavus
parent dd80b04235
commit a1ce428004
1 changed files with 5 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
ggml/src/ggml-openvino/ggml-quants.cpp
View File

@ -15,8 +15,8 @@
void unpack_32_4(const uint8_t* data, uint8_t* dst) {
std::fill_n(dst, 16, 0);
for (int j = 0; j < 16; ++j) {
uint8_t x = (data[j + 2] & 0x0F); // j+2 to skip scale bytes.
uint8_t y = (data[j + 2] >> 4);
uint8_t x = (data[j] & 0x0F);
uint8_t y = (data[j] >> 4);
if (j % 2 != 0) {
x <<= 4;
y <<= 4;
@ -41,7 +41,7 @@ void extract_q4_0_data(const ggml_tensor* tensor,
ov::parallel_for(scales_arr.get_size(), [&](size_t i) {
scales[i] = ov::float16::from_bits(*((uint16_t*)(data + i * bytes_per_block)));
biases[i] = ov::float16(-8.f * static_cast<float>(scales[i]));
unpack_32_4(data + i * bytes_per_block, weights + i * 16);
unpack_32_4(data + i * bytes_per_block + 2, weights + i * 16);
});
}
@ -58,8 +58,8 @@ void extract_q4_1_data(const ggml_tensor* tensor,
auto biases = biases_arr.data<ov::element_type_traits<ov::element::f16>::value_type>();
ov::parallel_for(scales_arr.get_size(), [&](size_t i) {
scales[i] = ov::float16::from_bits(*((uint16_t*)(data + i * bytes_per_block)));
biases[i] = ov::float16::from_bits(*((uint16_t*)(data + i * bytes_per_block + 1)));
unpack_32_4(data + i * bytes_per_block, weights + i * 16);
biases[i] = ov::float16::from_bits(*((uint16_t*)(data + i * bytes_per_block + 2)));
unpack_32_4(data + i * bytes_per_block + 4, weights + i * 16);
});
}