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

fix mmap usage when using host buffers

Browse source
This commit is contained in:
slaren 2024-10-07 20:42:52 +02:00
parent 0f3e091f1d
commit 59ee00a880
1 changed files with 2 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

3
src/llama.cpp
View file

@ -8903,7 +8903,8 @@ static bool llm_load_tensors(
bufs.reserve(n_max_backend_buffer);
// check if this backend device supports buffer_from_host_ptr
ggml_backend_dev_t dev = ggml_backend_buft_get_device(buft);
// when using a host buffer as the CPU bakcend buffer, use the CPU device to prioritize using buffer_from_host_ptr over the host buffer
ggml_backend_dev_t dev = ggml_backend_buft_get_device(buft == llama_default_buffer_type_cpu(model, true) ? ggml_backend_cpu_buffer_type() : buft);
bool buffer_from_host_ptr_supported = false;
if (dev) {
ggml_backend_dev_props props;