Lumpiasty/llama.cpp
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

llama: add default load-mode auto, which avoids mmap on iGPUs (#26081)

Browse Source 
* llama: add new default load-mode auto which picks mmap unless a non-Metal iGPU is used

* Update ggml/src/ggml-hexagon/ggml-hexagon.cpp

Co-authored-by: Max Krasnyansky <maxk@qti.qualcomm.com>

* set mmap_support to false on OpenCL backend

* fix order of load modes

* use -1 for auto

* resolve load mode auto earlier to correctly pick gpu host or cpu memory

* add load mode auto to llama-bench

* bump virtgpu api version, regenerate docs

---------

Co-authored-by: Piotr Wilkin (ilintar) <piotr.wilkin@syndatis.com>
Co-authored-by: Max Krasnyansky <maxk@qti.qualcomm.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
This commit is contained in:
Ruben Ortlam
2026-08-11 09:20:46 +03:00
committed by GitHub
co-authored by Max Krasnyansky Piotr Wilkin Georgi Gerganov
parent b3df57286c
commit 153d324bcf
33 changed files with 75 additions and 29 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/llama-model-loader.cpp
+1 -1
View File
@@ -543,7 +543,7 @@ llama_model_loader::llama_model_loader(
tensor_buft_overrides = param_tensor_buft_overrides_p;
this->use_mmap = load_mode == LLAMA_LOAD_MODE_MMAP || load_mode == LLAMA_LOAD_MODE_MMAP_MLOCK;
this->use_mmap = load_mode == LLAMA_LOAD_MODE_MMAP || load_mode == LLAMA_LOAD_MODE_MMAP_MLOCK || load_mode == LLAMA_LOAD_MODE_AUTO;
this->use_direct_io = load_mode == LLAMA_LOAD_MODE_DIRECT_IO;
if (!fname.empty()) {
src/llama-model.cpp
+17 -2
View File
@@ -1271,8 +1271,23 @@ bool llama_model_base::load_tensors(llama_model_loader & ml) {
this->ml = &ml; // to be used by create_tensor() and load_arch_tensors()
if (ml.use_mmap && params.load_mode == LLAMA_LOAD_MODE_AUTO) {
for (const auto & dev : devices) {
ggml_backend_dev_props props;
ggml_backend_dev_get_props(dev.dev, &props);
if (!props.caps.mmap_support) {
ml.use_mmap = false;
break;
}
}
}
const char * load_mode_name = params.load_mode == LLAMA_LOAD_MODE_AUTO
? llama_load_mode_name(ml.use_mmap ? LLAMA_LOAD_MODE_MMAP : LLAMA_LOAD_MODE_NONE)
: llama_load_mode_name(params.load_mode);
LLAMA_LOG_INFO("%s: loading model tensors, this can take a while... (load_mode = %s)\n",
__func__, llama_load_mode_name(params.load_mode));
__func__, load_mode_name);
// build a list of buffer types for the CPU and GPU devices
pimpl->cpu_buft_list = make_cpu_buft_list(devices, params.use_extra_bufts, params.no_host);
@@ -2452,7 +2467,7 @@ llama_model_params llama_model_default_params() {
/*.tensor_buft_overrides =*/ nullptr,
/*.n_gpu_layers =*/ -1,
/*.split_mode =*/ LLAMA_SPLIT_MODE_LAYER,
/*.load_mode =*/ LLAMA_LOAD_MODE_MMAP,
/*.load_mode =*/ LLAMA_LOAD_MODE_AUTO,
/*.main_gpu =*/ 0,
/*.tensor_split =*/ nullptr,
/*.progress_callback =*/ nullptr,
src/llama.cpp
+7 -4
View File
@@ -48,6 +48,8 @@ const char * llama_flash_attn_type_name(enum llama_flash_attn_type flash_attn_ty
const char * llama_load_mode_name(enum llama_load_mode load_mode) {
switch (load_mode) {
case LLAMA_LOAD_MODE_AUTO:
return "auto";
case LLAMA_LOAD_MODE_NONE:
return "none";
case LLAMA_LOAD_MODE_MMAP:
@@ -63,11 +65,12 @@ const char * llama_load_mode_name(enum llama_load_mode load_mode) {
}
enum llama_load_mode llama_load_mode_from_str(const char * str) {
if (std::strcmp(str, "none") == 0) { return LLAMA_LOAD_MODE_NONE; }
if (std::strcmp(str, "mmap") == 0) { return LLAMA_LOAD_MODE_MMAP; }
if (std::strcmp(str, "mlock") == 0) { return LLAMA_LOAD_MODE_MLOCK; }
if (std::strcmp(str, "auto") == 0) { return LLAMA_LOAD_MODE_AUTO; }
if (std::strcmp(str, "none") == 0) { return LLAMA_LOAD_MODE_NONE; }
if (std::strcmp(str, "mmap") == 0) { return LLAMA_LOAD_MODE_MMAP; }
if (std::strcmp(str, "mlock") == 0) { return LLAMA_LOAD_MODE_MLOCK; }
if (std::strcmp(str, "mmap+mlock") == 0) { return LLAMA_LOAD_MODE_MMAP_MLOCK; }
if (std::strcmp(str, "dio") == 0) { return LLAMA_LOAD_MODE_DIRECT_IO; }
if (std::strcmp(str, "dio") == 0) { return LLAMA_LOAD_MODE_DIRECT_IO; }
throw std::invalid_argument(std::string("unknown load mode: ") + str);
}