tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-08-05 08:28:37 -04:00
Code Issues Packages Projects Releases Wiki Activity

fix: Fix resize vs reserve and skip null tensors in size computation

Browse Source 
https://github.com/ggml-org/llama.cpp/pull/13979/files#r2149469788

Branch: HybridRecurrentCache

Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
Co-Authored-By: @younesbelkada
This commit is contained in:
Gabe Goodhart
2025-06-16 13:34:25 -06:00
parent 11cd80d5de
commit 9db44a2a63
1 changed files with 8 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
src/llama-kv-cache-recurrent.cpp
View File

@@ -60,8 +60,8 @@ llama_kv_cache_recurrent::llama_kv_cache_recurrent(
return it->second;
};
k_l.reserve(n_layer);
v_l.reserve(n_layer);
k_l.resize(n_layer);
v_l.resize(n_layer);
for (int i = 0; i < n_layer; i++) {
if (filter && !filter(i)) {
@@ -647,7 +647,9 @@ size_t llama_kv_cache_recurrent::size_k_bytes() const {
size_t size_k_bytes = 0;
for (const auto & k : k_l) {
size_k_bytes += ggml_nbytes(k);
if (k != nullptr) {
size_k_bytes += ggml_nbytes(k);
}
}
return size_k_bytes;
@@ -657,7 +659,9 @@ size_t llama_kv_cache_recurrent::size_v_bytes() const {
size_t size_v_bytes = 0;
for (const auto & v : v_l) {
size_v_bytes += ggml_nbytes(v);
if (v != nullptr) {
size_v_bytes += ggml_nbytes(v);
}
}
return size_v_bytes;