root/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-11-14 23:09:53 +00:00
Code Issues Actions 14 Packages Projects Releases Wiki Activity

Fixed dequant precision issues in Q4_1 and Q5_1 (#9711)

Browse Source
This commit is contained in:
Ouadie EL FAROUKI 2024-10-03 07:50:44 +01:00 committed by GitHub
parent c83ad6d01e
commit 5639971466
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 8 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
ggml/src/ggml-sycl/dequantize.hpp
View File

@ -55,12 +55,12 @@ static __dpct_inline__ void dequantize_q4_1(const void *vx, const int64_t ib,
#ifdef GGML_SYCL_F16 #ifdef GGML_SYCL_F16
// v = v * {d, d}; // v = v * {d, d};
// v = v + {m, m}; // v = v + {m, m};
v.s0() = (v.s0() * d) + m; v.s0() = sycl::fma(v.s0(), d, m);
v.s1() = (v.s1() * d) + m; v.s1() = sycl::fma(v.s1(), d, m);
#else #else
v.x() = (v.x() * d) + m; v.x() = sycl::fma(v.x(), d, m);
v.y() = (v.y() * d) + m; v.y() = sycl::fma(v.y(), d, m);
#endif // GGML_SYCL_F16 #endif // GGML_SYCL_F16
} }
@ -110,11 +110,11 @@ static __dpct_inline__ void dequantize_q5_1(const void *vx, const int64_t ib,
#ifdef GGML_SYCL_F16 #ifdef GGML_SYCL_F16
// v = v * {d, d}; // v = v * {d, d};
// v = v + {m, m}; // v = v + {m, m};
v.s0() = (v.s0() * d) + m; v.s0() = sycl::fma(v.s0(), d, m);
v.s1() = (v.s1() * d) + m; v.s1() = sycl::fma(v.s1(), d, m);
#else #else
v.x() = (v.x() * d) + m; v.x() = sycl::fma(v.x(), d, m);
v.y() = (v.y() * d) + m; v.y() = sycl::fma(v.y(), d, m);
#endif // GGML_SYCL_F16 #endif // GGML_SYCL_F16
} }