darthrubik/llama-cpp
1
0
Fork 0
Code Issues Pull Requests 1 Packages Projects Releases Wiki Activity

Fix !30793
Copilot Setup Steps / copilot-setup-steps (push) Has been cancelled
Check Pre-Tokenizer Hashes / pre-tokenizer-hashes (push) Has been cancelled
Python check requirements.txt / check-requirements (push) Has been cancelled
Python Type-Check / python type-check (push) Has been cancelled
Update Operations Documentation / update-ops-docs (push) Has been cancelled

Browse Source
This commit is contained in:
Gaurav Garg
2026-10-22 19:30:57 +05:30
committed by GitHub
parent 5c19a5a756
commit 36cbf415f1
1 changed files with 14 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/llama-model.cpp
+14 -6
View File
@@ -1968,12 +1968,20 @@ bool llama_model_base::load_tensors(llama_model_loader & ml) {
if (buf == nullptr) {
throw std::runtime_error(format("unable to allocate %s buffer", ggml_backend_buft_name(buft)));
}
// a multi_buffer has no base to lock, its parts are usually pinned memory (host buffer types with a max size)
if (use_mlock && ggml_backend_buffer_is_host(buf) && ggml_backend_buffer_get_base(buf) != nullptr) {
pimpl->mlock_bufs.emplace_back(new llama_mlock);
auto & mlock_buf = pimpl->mlock_bufs.back();
mlock_buf->init (ggml_backend_buffer_get_base(buf));
mlock_buf->grow_to(ggml_backend_buffer_get_size(buf));
// lock the buffers that hold the tensors: a multi_buffer has no base, and its parts are pageable memory when pinning fails
if (use_mlock && !ml.no_alloc && ggml_backend_buffer_is_host(buf)) {
std::vector<ggml_backend_buffer_t> lock_bufs;
for (ggml_tensor * t = ggml_get_first_tensor(ctx); t != nullptr; t = ggml_get_next_tensor(ctx, t)) {
if (t->buffer != nullptr && std::find(lock_bufs.begin(), lock_bufs.end(), t->buffer) == lock_bufs.end()) {
lock_bufs.push_back(t->buffer);
}
}
for (ggml_backend_buffer_t lock_buf : lock_bufs) {
pimpl->mlock_bufs.emplace_back(new llama_mlock);
auto & mlock_buf = pimpl->mlock_bufs.back();
mlock_buf->init (ggml_backend_buffer_get_base(lock_buf));
mlock_buf->grow_to(ggml_backend_buffer_get_size(lock_buf));
}
}
bufs.emplace_back(buf);
}