mirror of
https://github.com/ggml-org/llama.cpp.git
synced 2025-10-29 08:41:22 +00:00
cuda: get_rows: dfloat2 -> float2
This commit is contained in:
@@ -30,7 +30,7 @@ static __global__ void k_get_rows(
|
|||||||
const int y_offset = qr == 1 ? 1 : qk/2;
|
const int y_offset = qr == 1 ? 1 : qk/2;
|
||||||
|
|
||||||
// dequantize
|
// dequantize
|
||||||
dfloat2 v;
|
float2 v;
|
||||||
dequantize_kernel(src0_row, ib, iqs, v);
|
dequantize_kernel(src0_row, ib, iqs, v);
|
||||||
|
|
||||||
dst_row[iybs + iqs + 0] = ggml_cuda_cast<dst_t>(v.x);
|
dst_row[iybs + iqs + 0] = ggml_cuda_cast<dst_t>(v.x);
|
||||||
|
|||||||
Reference in New Issue
Block a user