vrr/prima.cpp
2
0
Fork 0
mirror of https://github.com/Lizonghang/prima.cpp.git synced 2025-09-06 15:49:03 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

skip tensors on CUDA in manage_graph_tensors

Browse source
This commit is contained in:
Zonghang Li 2025-02-11 09:49:17 +04:00
parent c4c6a642fc
commit 261c88f058
1 changed files with 9 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

9
src/llama.cpp
View file

@ -17776,10 +17776,19 @@ static void manage_graph_tensors(struct ggml_cgraph * cgraph, int advice, bool f
for (int i = 0; i < ggml_graph_n_leafs(cgraph); i++) { for (int i = 0; i < ggml_graph_n_leafs(cgraph); i++) {
struct ggml_tensor * cur = ggml_graph_leaf(cgraph, i); struct ggml_tensor * cur = ggml_graph_leaf(cgraph, i);
if (strstr(cur->name, "weight") == nullptr || cur->data == nullptr) { if (strstr(cur->name, "weight") == nullptr || cur->data == nullptr) {
continue; continue;
} }
const char * backend_name = ggml_backend_buffer_name(cur->buffer);
if (backend_name) {
std::string lower_name(backend_name);
std::transform(lower_name.begin(), lower_name.end(), lower_name.begin(),
[](unsigned char c) { return std::tolower(c); });
if (lower_name.find("cuda") != std::string::npos) continue;
}
size_t size = ggml_nbytes(cur); size_t size = ggml_nbytes(cur);
size_t first = reinterpret_cast<size_t>(cur->data); size_t first = reinterpret_cast<size_t>(cur->data);
size_t last = first + size; size_t last = first + size;