Spaces:
Running
Running
Commit
·
2746afd
1
Parent(s):
88864af
CUDA: fix compilation with GGML_CUDA_F16 (llama/14837)
Browse files
ggml/src/ggml-cuda/convert.cu
CHANGED
|
@@ -31,8 +31,8 @@ static __global__ void dequantize_block(const void * __restrict__ vx, dst_t * __
|
|
| 31 |
dequantize_kernel(vx, ib, iqs, v);
|
| 32 |
|
| 33 |
const int64_t iy0 = ((i03*ne02 + i02)*ne01 + i01)*ne00 + iybs + iqs;
|
| 34 |
-
y[iy0 + 0] = v.x;
|
| 35 |
-
y[iy0 + y_offset] = v.y;
|
| 36 |
}
|
| 37 |
|
| 38 |
template <bool need_check>
|
|
|
|
| 31 |
dequantize_kernel(vx, ib, iqs, v);
|
| 32 |
|
| 33 |
const int64_t iy0 = ((i03*ne02 + i02)*ne01 + i01)*ne00 + iybs + iqs;
|
| 34 |
+
y[iy0 + 0] = float(v.x);
|
| 35 |
+
y[iy0 + y_offset] = float(v.y);
|
| 36 |
}
|
| 37 |
|
| 38 |
template <bool need_check>
|