CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-27 08:21:30 +00:00
Code Issues Packages Projects Releases Wiki Activity

CUDA: fixed row rounding for 0 tensor splits (#4594)

Browse Source
This commit is contained in:
Johannes Gäßler
2023-12-23 09:16:33 +01:00
committed by GitHub
parent 7082d24cec
commit e0a4002273
1 changed files with 6 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
ggml-cuda.cu
View File

@@ -7937,12 +7937,16 @@ static void ggml_cuda_op_mul_mat(
if (id != 0) {
row_low[id] = ne01*g_tensor_split[id];
row_low[id] -= row_low[id] % rounding;
if (row_low[id] < ne01) {
row_low[id] -= row_low[id] % rounding;
}
}
if (id != g_device_count - 1) {
row_high[id] = ne01*g_tensor_split[id + 1];
row_high[id] -= row_high[id] % rounding;
if (row_high[id] < ne01) {
row_high[id] -= row_high[id] % rounding;
}
}
}
}