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

rpc : early register backend devices (#11262)

Browse source 
Early register RPC devices and do not propagate RPC specifics in the
llama model structures.

ref: #10609
This commit is contained in:
Radoslav Gerganov 2025-01-17 10:57:09 +02:00 • committed by GitHub
parent a133566d34
commit 667d72846c
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
10 changed files with 61 additions and 55 deletions
Download patch file Download diff file Expand all files Collapse all files

1
src/llama-model.cpp
View file

@ -3717,7 +3717,6 @@ struct llama_model_params llama_model_default_params() {
/*.split_mode =*/ LLAMA_SPLIT_MODE_LAYER,
/*.main_gpu =*/ 0,
/*.tensor_split =*/ nullptr,
/*.rpc_servers =*/ nullptr,
/*.progress_callback =*/ nullptr,
/*.progress_callback_user_data =*/ nullptr,
/*.kv_overrides =*/ nullptr,

2
src/llama-model.h
View file

@ -323,8 +323,6 @@ struct llama_model {
// gguf metadata
std::unordered_map<std::string, std::string> gguf_kv;
std::vector<std::string> rpc_servers;
// list of devices used in this model
std::vector<ggml_backend_dev_t> devices;

41
src/llama.cpp
View file

@ -9399,47 +9399,6 @@ static struct llama_model * llama_model_load_from_file_impl(
};
}
if (params.rpc_servers != nullptr && params.rpc_servers[0] != '\0') {
// split the servers set them into model->rpc_servers
std::string servers(params.rpc_servers);
size_t pos = 0;
while ((pos = servers.find(',')) != std::string::npos) {
std::string server = servers.substr(0, pos);
model->rpc_servers.push_back(server);
servers.erase(0, pos + 1);
}
model->rpc_servers.push_back(servers);
}
// add RPC devices
if (!model->rpc_servers.empty()) {
ggml_backend_reg_t rpc_reg = ggml_backend_reg_by_name("RPC");
if (!rpc_reg) {
LLAMA_LOG_ERROR("%s: failed to find RPC backend\n", __func__);
llama_model_free(model);
return nullptr;
}
typedef ggml_backend_dev_t (*ggml_backend_rpc_add_device_t)(const char * endpoint);
ggml_backend_rpc_add_device_t ggml_backend_rpc_add_device_fn = (ggml_backend_rpc_add_device_t) ggml_backend_reg_get_proc_address(rpc_reg, "ggml_backend_rpc_add_device");
if (!ggml_backend_rpc_add_device_fn) {
LLAMA_LOG_ERROR("%s: failed to find RPC device add function\n", __func__);
llama_model_free(model);
return nullptr;
}
for (const std::string & server : model->rpc_servers) {
ggml_backend_dev_t dev = ggml_backend_rpc_add_device_fn(server.c_str());
if (dev) {
model->devices.push_back(dev);
} else {
LLAMA_LOG_ERROR("%s: failed to add RPC device for server '%s'\n", __func__, server.c_str());
llama_model_free(model);
return nullptr;
}
}
}
// create list of devices to use with this model
if (params.devices) {
for (ggml_backend_dev_t * dev = params.devices; *dev; ++dev) {