vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Apply suggestions from code review

Browse source 
Co-authored-by: slaren <slarengh@gmail.com>
This commit is contained in:
Markus Tavenrath 2024-06-13 15:55:23 +02:00 • committed by GitHub
parent 86869fbdab
commit c39d5ecd2b
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 2 additions and 4 deletions
Download patch file Download diff file Expand all files Collapse all files

6
llama.cpp
View file

@ -3851,7 +3851,7 @@ struct llama_model_loader {
size_t buffer_idx = 0; // buffer to use for async loads
ggml_backend_t cuda_backend = nullptr;
if (!use_mmap) {
if (!use_mmap && !check_tensors) {
// When not using mmaped io use async uploads from pinned memory to GPU memory.
// First determine if the CUDA backend is active, and if so, determine the device ID.
ggml_backend_buffer_t buf = bufs_mmap.count(0) ? bufs_mmap.at(0) : nullptr;
@ -3939,8 +3939,7 @@ struct llama_model_loader {
size_t bytes_read = 0;
while (bytes_read < n_size)
{
while (bytes_read < n_size) {
size_t read_iteration = std::min<size_t>(buffer_size, n_size - bytes_read);
ggml_backend_event_synchronize(events[buffer_idx]);
@ -3982,7 +3981,6 @@ struct llama_model_loader {
}
#endif
// check validation results
bool validation_failed = false;
for (auto & future : validation_result) {