happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

llama : allow using iGPUs with --device (#15951) 

* llama : allow using iGPUs with --device

* mtmd : allow iGPU

* rpc-server : allow iGPU
This commit is contained in:
Diego Devesa 2025-09-13 07:49:49 -07:00 committed by GitHub
parent 55758b00ca
commit 50f4281a6f
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
3 changed files with 8 additions and 24 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

21
common/arg.cpp
View File

@ -1304,7 +1304,7 @@ static std::vector<ggml_backend_dev_t> parse_device_list(const std::string & val
} else { } else {
for (const auto & device : dev_names) { for (const auto & device : dev_names) {
auto * dev = ggml_backend_dev_by_name(device.c_str()); auto * dev = ggml_backend_dev_by_name(device.c_str());
if (!dev || ggml_backend_dev_type(dev) != GGML_BACKEND_DEVICE_TYPE_GPU) { if (!dev || ggml_backend_dev_type(dev) == GGML_BACKEND_DEVICE_TYPE_CPU) {
throw std::invalid_argument(string_format("invalid device: %s", device.c_str())); throw std::invalid_argument(string_format("invalid device: %s", device.c_str()));
} }
devices.push_back(dev); devices.push_back(dev);
@ -1314,7 +1314,7 @@ static std::vector<ggml_backend_dev_t> parse_device_list(const std::string & val
return devices; return devices;
} }
static void add_rpc_devices(std::string servers) { static void add_rpc_devices(const std::string & servers) {
auto rpc_servers = string_split<std::string>(servers, ','); auto rpc_servers = string_split<std::string>(servers, ',');
if (rpc_servers.empty()) { if (rpc_servers.empty()) {
throw std::invalid_argument("no RPC servers specified"); throw std::invalid_argument("no RPC servers specified");
@ -2516,24 +2516,15 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
{"--list-devices"}, {"--list-devices"},
"print list of available devices and exit", "print list of available devices and exit",
[](common_params &) { [](common_params &) {
std::vector<ggml_backend_dev_t> rpc_devices; std::vector<ggml_backend_dev_t> devices;
std::vector<ggml_backend_dev_t> all_devices;
for (size_t i = 0; i < ggml_backend_dev_count(); ++i) { for (size_t i = 0; i < ggml_backend_dev_count(); ++i) {
auto * dev = ggml_backend_dev_get(i); auto * dev = ggml_backend_dev_get(i);
if (ggml_backend_dev_type(dev) == GGML_BACKEND_DEVICE_TYPE_GPU) { if (ggml_backend_dev_type(dev) != GGML_BACKEND_DEVICE_TYPE_CPU) {
ggml_backend_reg_t reg = ggml_backend_dev_backend_reg(dev); devices.push_back(dev);
if (ggml_backend_reg_name(reg) == std::string("RPC")) {
rpc_devices.push_back(dev);
} else {
all_devices.push_back(dev);
}
} }
} }
// insert RPC devices in front
all_devices.insert(all_devices.begin(), rpc_devices.begin(), rpc_devices.end());
printf("Available devices:\n"); printf("Available devices:\n");
for (size_t i = 0; i < all_devices.size(); ++i) { for (auto * dev : devices) {
auto * dev = all_devices[i];
size_t free, total; size_t free, total;
ggml_backend_dev_memory(dev, &free, &total); ggml_backend_dev_memory(dev, &free, &total);
printf(" %s: %s (%zu MiB, %zu MiB free)\n", ggml_backend_dev_name(dev), ggml_backend_dev_description(dev), total / 1024 / 1024, free / 1024 / 1024); printf(" %s: %s (%zu MiB, %zu MiB free)\n", ggml_backend_dev_name(dev), ggml_backend_dev_description(dev), total / 1024 / 1024, free / 1024 / 1024);

1
tools/mtmd/clip.cpp
View File

@ -406,6 +406,7 @@ struct clip_ctx {
} }
if (!backend) { if (!backend) {
backend = ggml_backend_init_by_type(GGML_BACKEND_DEVICE_TYPE_GPU, nullptr); backend = ggml_backend_init_by_type(GGML_BACKEND_DEVICE_TYPE_GPU, nullptr);
backend = backend ? backend : ggml_backend_init_by_type(GGML_BACKEND_DEVICE_TYPE_IGPU, nullptr);
} }
} }

10
tools/rpc/rpc-server.cpp
View File

@ -227,15 +227,7 @@ static ggml_backend_t create_backend(const rpc_server_params & params) {
} }
} }
// try to initialize a GPU backend first backend = ggml_backend_init_best();
if (!backend) {
backend = ggml_backend_init_by_type(GGML_BACKEND_DEVICE_TYPE_GPU, nullptr);
}
// if there aren't GPU backends fallback to CPU backend
if (!backend) {
backend = ggml_backend_init_by_type(GGML_BACKEND_DEVICE_TYPE_CPU, nullptr);
}
if (backend) { if (backend) {
fprintf(stderr, "%s: using %s backend\n", __func__, ggml_backend_name(backend)); fprintf(stderr, "%s: using %s backend\n", __func__, ggml_backend_name(backend));