ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-03-13 15:30:03 +00:00
Code Issues Packages Projects Releases Wiki Activity

DeepSeek imatrix stuff (#250)

Browse Source 
* This gives us ~20% TG speedup for DeepSeek on CUDA

* Slightly better

* Also do it for plain (not fused) mul_mat_id

* Guard against numerical precision issues for MLA on CUDA

* imatrix: wv_b <-> wkv_b

---------

Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
This commit is contained in:
Kawrakow
2025-03-10 16:19:09 +02:00
committed by GitHub
parent 699c9cb7f6
commit a48e163247
4 changed files with 21 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
examples/imatrix/imatrix.cpp
View File

@@ -195,7 +195,7 @@ bool IMatrixCollector::collect_imatrix(struct ggml_tensor * t, bool ask, void *
if (m_params.verbosity > 1) {
printf("%s[%d]: %32s, %s, %5d x %5d, %d\n", __func__, m_last_call, wname.c_str(), ggml_op_name(t->op), (int)src1->ne[0], (int)src1->ne[1], (int)src1->type);
}
for (int row = 0; row < (int)src1->ne[1]; ++row) {
for (int row = 0; row < (int)(src1->ne[1]*src1->ne[2]); ++row) {
const float * x = data + row * src1->ne[0];
for (int j = 0; j < (int)src1->ne[0]; ++j) {
e.values[j] += x[j]*x[j];