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

server : add lora hotswap endpoint

Browse source
This commit is contained in:
ngxson 2024-08-04 19:46:13 +02:00
parent ecf6b7f23e
commit e5c2d16739
3 changed files with 109 additions and 21 deletions
Download patch file Download diff file Expand all files Collapse all files

53
common/common.cpp
View file

@ -684,14 +684,22 @@ bool gpt_params_find_arg(int argc, char ** argv, const std::string & arg, gpt_pa
}
if (arg == "--lora") {
CHECK_ARG
params.lora_adapter.emplace_back(argv[i], 1.0f);
params.lora_adapters.push_back({
std::string(argv[i]),
1.0,
nullptr,
});
return true;
}
if (arg == "--lora-scaled") {
CHECK_ARG
const char* lora_adapter = argv[i];
std::string lora_adapter = argv[i];
CHECK_ARG
params.lora_adapter.emplace_back(lora_adapter, std::stof(argv[i]));
params.lora_adapters.push_back({
lora_adapter,
std::stof(argv[i]),
nullptr,
});
return true;
}
if (arg == "--control-vector") {
@ -1654,6 +1662,7 @@ void gpt_params_print_usage(int /*argc*/, char ** argv, const gpt_params & param
"https://github.com/ggerganov/llama.cpp/wiki/Templates-supported-by-llama_chat_apply_template" });
options.push_back({ "server", "-sps, --slot-prompt-similarity SIMILARITY",
"how much the prompt of a request must match the prompt of a slot in order to use that slot (default: %.2f, 0.0 = disabled)\n", params.slot_prompt_similarity });
options.push_back({ "server", "--lora-no-apply", "only load LoRA adapters without applying them (apply later via POST /lora-adapters) (default: %s)", params.lora_no_apply ? "enabled" : "disabled"});
#ifndef LOG_DISABLE_LOGS
options.push_back({ "logging" });
@ -2091,17 +2100,18 @@ std::tuple<struct llama_model *, struct llama_context *> llama_init_from_gpt_par
}
}
for (unsigned int i = 0; i < params.lora_adapter.size(); ++i) {
const std::string & lora_adapter = std::get<0>(params.lora_adapter[i]);
float lora_scale = std::get<1>(params.lora_adapter[i]);
auto adapter = llama_lora_adapter_init(model, lora_adapter.c_str());
if (adapter == nullptr) {
fprintf(stderr, "%s: error: failed to apply lora adapter\n", __func__);
// load and optionally apply lora adapters
for (auto & la : params.lora_adapters) {
la.adapter = llama_lora_adapter_init(model, la.path.c_str());
if (la.adapter == nullptr) {
fprintf(stderr, "%s: error: failed to apply lora adapter '%s'\n", __func__, la.path.c_str());
llama_free(lctx);
llama_free_model(model);
return std::make_tuple(nullptr, nullptr);
}
llama_lora_adapter_set(lctx, adapter, lora_scale);
}
if (!params.lora_no_apply) {
llama_lora_adapters_apply(lctx, params.lora_adapters);
}
if (params.ignore_eos) {
@ -2138,6 +2148,15 @@ std::tuple<struct llama_model *, struct llama_context *> llama_init_from_gpt_par
return std::make_tuple(model, lctx);
}
void llama_lora_adapters_apply(struct llama_context * ctx, std::vector<llama_lora_adapter_container> & lora_adapters) {
llama_lora_adapter_clear(ctx);
for (auto & la : lora_adapters) {
if (la.scale != 0.0f) {
llama_lora_adapter_set(ctx, la.adapter, la.scale);
}
}
}
struct llama_model_params llama_model_params_from_gpt_params(const gpt_params & params) {
auto mparams = llama_model_default_params();
@ -3160,18 +3179,16 @@ void yaml_dump_non_result_info(FILE * stream, const gpt_params & params, const l
}
fprintf(stream, "lora:\n");
for (std::tuple<std::string, float> la : params.lora_adapter) {
if (std::get<1>(la) != 1.0f) {
continue;
for (auto & la : params.lora_adapters) {
if (la.scale == 1.0f) {
fprintf(stream, " - %s\n", la.path.c_str());
}
fprintf(stream, " - %s\n", std::get<0>(la).c_str());
}
fprintf(stream, "lora_scaled:\n");
for (std::tuple<std::string, float> la : params.lora_adapter) {
if (std::get<1>(la) == 1.0f) {
continue;
for (auto & la : params.lora_adapters) {
if (la.scale != 1.0f) {
fprintf(stream, " - %s: %f\n", la.path.c_str(), la.scale);
}
fprintf(stream, " - %s: %f\n", std::get<0>(la).c_str(), std::get<1>(la));
}
fprintf(stream, "main_gpu: %d # default: 0\n", params.main_gpu);
fprintf(stream, "min_keep: %d # default: 0 (disabled)\n", sparams.min_keep);

13
common/common.h
View file

@ -33,6 +33,12 @@
#define DEFAULT_MODEL_PATH "models/7B/ggml-model-f16.gguf"
struct llama_lora_adapter_container {
std::string path;
float scale;
struct llama_lora_adapter * adapter;
};
// build info
extern int LLAMA_BUILD_NUMBER;
extern char const * LLAMA_COMMIT;
@ -126,8 +132,8 @@ struct gpt_params {
std::vector<std::string> antiprompt; // strings upon which more user input is prompted (a.k.a. reverse prompts)
std::vector<llama_model_kv_override> kv_overrides;
// TODO: avoid tuple, use struct
std::vector<std::tuple<std::string, float>> lora_adapter; // lora adapter path with user defined scale
bool lora_no_apply = false; // only load lora to memory, but do not apply it to ctx (user can manually apply lora later using llama_lora_adapter_apply)
std::vector<llama_lora_adapter_container> lora_adapters; // lora adapter path with user defined scale
std::vector<llama_control_vector_load_info> control_vectors; // control vector with user defined scale
@ -317,6 +323,9 @@ struct llama_context_params llama_context_params_from_gpt_params(const gpt_param
struct llama_model * llama_load_model_from_url(const char * model_url, const char * path_model, const char * hf_token, const struct llama_model_params & params);
struct llama_model * llama_load_model_from_hf(const char * repo, const char * file, const char * path_model, const char * hf_token, const struct llama_model_params & params);
// clear LoRA adapters from context, then apply new list of adapters
void llama_lora_adapters_apply(struct llama_context * ctx, std::vector<llama_lora_adapter_container> & lora_adapters);
// Batch utils
void llama_batch_clear(struct llama_batch & batch);

64
examples/server/server.cpp
View file

@ -78,6 +78,7 @@ enum server_task_type {
SERVER_TASK_TYPE_SLOT_SAVE,
SERVER_TASK_TYPE_SLOT_RESTORE,
SERVER_TASK_TYPE_SLOT_ERASE,
SERVER_TASK_TYPE_SET_LORA,
};
struct server_task {
@ -1847,6 +1848,14 @@ struct server_context {
};
queue_results.send(result);
} break;
case SERVER_TASK_TYPE_SET_LORA:
{
llama_lora_adapters_apply(ctx, params.lora_adapters);
server_task_result result;
result.id = task.id;
result.data = json{{ "success", true }};
queue_results.send(result);
} break;
}
}
@ -3325,6 +3334,55 @@ int main(int argc, char ** argv) {
return res.set_content(root.dump(), "application/json; charset=utf-8");
};
const auto handle_lora_adapters_list = [&](const httplib::Request & req, httplib::Response & res) {
res.set_header("Access-Control-Allow-Origin", req.get_header_value("Origin"));
json result = json::array();
for (size_t i = 0; i < ctx_server.params.lora_adapters.size(); ++i) {
auto & la = ctx_server.params.lora_adapters[i];
result.push_back({
{"id", i},
{"path", la.path},
{"scale", la.scale},
});
}
res.set_content(result.dump(), "application/json");
res.status = 200; // HTTP OK
};
const auto handle_lora_adapters_apply = [&](const httplib::Request & req, httplib::Response & res) {
res.set_header("Access-Control-Allow-Origin", req.get_header_value("Origin"));
const std::vector<json> body = json::parse(req.body);
int max_idx = ctx_server.params.lora_adapters.size();
// clear existing value
for (auto & la : ctx_server.params.lora_adapters) {
la.scale = 0.0f;
}
// set value
for (auto entry : body) {
int id = entry.at("id");
float scale = entry.at("scale");
if (0 <= id && id < max_idx) {
ctx_server.params.lora_adapters[id].scale = scale;
} else {
throw std::runtime_error("invalid adapter id");
}
}
server_task task;
task.type = SERVER_TASK_TYPE_SET_LORA;
const int id_task = ctx_server.queue_tasks.post(task);
ctx_server.queue_results.add_waiting_task_id(id_task);
server_task_result result = ctx_server.queue_results.recv(id_task);
ctx_server.queue_results.remove_waiting_task_id(id_task);
res.set_content(result.data.dump(), "application/json");
res.status = 200; // HTTP OK
};
auto handle_static_file = [](unsigned char * content, size_t len, const char * mime_type) {
return [content, len, mime_type](const httplib::Request &, httplib::Response & res) {
res.set_content(reinterpret_cast<const char*>(content), len, mime_type);
@ -3363,7 +3421,6 @@ int main(int argc, char ** argv) {
// register API routes
svr->Get ("/health", handle_health);
svr->Get ("/slots", handle_slots);
svr->Get ("/metrics", handle_metrics);
svr->Get ("/props", handle_props);
svr->Get ("/v1/models", handle_models);
@ -3378,6 +3435,11 @@ int main(int argc, char ** argv) {
svr->Post("/v1/embeddings", handle_embeddings);
svr->Post("/tokenize", handle_tokenize);
svr->Post("/detokenize", handle_detokenize);
// LoRA adapters hotswap
svr->Get ("/lora-adapters", handle_lora_adapters_list);
svr->Post("/lora-adapters", handle_lora_adapters_apply);
// Save & load slots
svr->Get ("/slots", handle_slots);
if (!params.slot_save_path.empty()) {
// only enable slot endpoints if slot_save_path is set
svr->Post("/slots/:id_slot", handle_slots_action);