CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-11-21 12:16:57 +00:00
Code Issues Packages Projects Releases Wiki Activity

ggml : Fix transposed SOLVE_TRI result (#17323)

Browse Source 
* Did someone transpose the SOLVE_TRI result matrix? Perhaps...

* Update ggml/src/ggml-cpu/ops.cpp

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>

* Update ggml/src/ggml-cpu/ops.cpp

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>

---------

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
This commit is contained in:
Piotr Wilkin (ilintar)
2025-11-20 11:58:21 +01:00
committed by GitHub
parent a7784a8b1d
commit 845f200b28
1 changed files with 2 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
ggml/src/ggml-cpu/ops.cpp
View File

@@ -9696,13 +9696,12 @@ static void ggml_compute_forward_solve_tri_f32(const struct ggml_compute_params
for (int64_t i00 = 0; i00 < n; ++i00) {
float sum = 0.0f;
for (int64_t t = 0; t < i00; ++t) {
sum += A_batch[i00 * n + t] * X_batch[i01 * n + t];
sum += A_batch[i00 * n + t] * X_batch[t * k + i01];
}
const float diag = A_batch[i00 * n + i00];
GGML_ASSERT(diag != 0.0f && "Zero diagonal in triangular matrix");
X_batch[i01 * n + i00] = (B_batch[i00 * k + i01] - sum) / diag;
X_batch[i00 * k + i01] = (B_batch[i00 * k + i01] - sum) / diag;
}
}
}