vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

addressed comment

Browse source
This commit is contained in:
Alan Gray 2024-10-10 06:05:12 -07:00
parent 95c8b9c1b7
commit d07dc44c63
1 changed files with 2 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

4
ggml/src/ggml-cuda/dmmv.cu
View file

@ -419,8 +419,8 @@ static __device__ void convert_f16(const void * vx, const int64_t ib, const int
// load 2 halfs into register in a single instruction
const half2 x_reg = *((half2 *) &(x[ib + iqs]));
// automatic half -> float type cast if dfloat == float
v.x = x_reg.x;
v.y = x_reg.y;
v.x = __low2float(x_reg);
v.y = __high2float(x_reg);
}
static constexpr __device__ dequantize_kernel_t get_dequantize_kernel(ggml_type type) {