matt/ollama37
1
0
Fork 0
mirror of https://github.com/dogkeeper886/ollama37.git synced 2025-12-10 15:57:04 +00:00
Code Issues Packages Projects Releases Wiki Activity

update llama.cpp submodule to ceca1ae (#3064)

Browse Source
This commit is contained in:
Jeffrey Morgan
2024-03-11 12:57:48 -07:00
committed by GitHub
parent f878e91070
commit 369eda65f5
6 changed files with 36 additions and 61 deletions
Download Patch File Download Diff File Expand all files Collapse all files

26
llm/ext_server/ext_server.cpp
View File

@@ -26,7 +26,7 @@
#endif // GGML_USE_CUBLAS
// Expose the llama server as a callable extern "C" API
server_context *llama = NULL;
llama_server_context *llama = NULL;
std::thread ext_server_thread;
bool shutting_down = false;
std::atomic_int recv_counter;
@@ -57,7 +57,7 @@ void llama_server_init(ext_server_params *sparams, ext_server_resp_t *err) {
err->id = 0;
err->msg[0] = '\0';
try {
llama = new server_context;
llama = new llama_server_context;
gpt_params params;
params.n_ctx = sparams->n_ctx;
params.n_batch = sparams->n_batch;
@@ -125,7 +125,7 @@ void llama_server_init(ext_server_params *sparams, ext_server_resp_t *err) {
return;
}
llama->init();
llama->initialize();
} catch (std::exception &e) {
err->id = -1;
snprintf(err->msg, err->msg_len, "exception %s", e.what());
@@ -144,13 +144,13 @@ void llama_server_start() {
LOG_TEE("llama server main loop starting\n");
ggml_time_init();
llama->queue_tasks.on_new_task(std::bind(
&server_context::process_single_task, llama, std::placeholders::_1));
&llama_server_context::process_single_task, llama, std::placeholders::_1));
llama->queue_tasks.on_finish_multitask(std::bind(
&server_context::on_finish_multitask, llama, std::placeholders::_1));
&llama_server_context::on_finish_multitask, llama, std::placeholders::_1));
llama->queue_tasks.on_run_slots(std::bind(
&server_context::update_slots, llama));
&llama_server_context::update_slots, llama));
llama->queue_results.on_multitask_update(std::bind(
&server_queue::update_multitask,
&llama_server_queue::update_multitask,
&llama->queue_tasks,
std::placeholders::_1,
std::placeholders::_2,
@@ -198,7 +198,7 @@ void llama_server_completion(const char *json_req, ext_server_resp_t *resp) {
json data = json::parse(json_req);
resp->id = llama->queue_tasks.get_new_id();
llama->queue_results.add_waiting_task_id(resp->id);
llama->request_completion(resp->id, -1, data, false, false);
llama->request_completion(resp->id, data, false, false, -1);
} catch (std::exception &e) {
snprintf(resp->msg, resp->msg_len, "exception %s", e.what());
} catch (...) {
@@ -216,9 +216,9 @@ void llama_server_completion_next_result(const int task_id,
std::string result_json;
try {
atomicRecv ar(recv_counter);
server_task_result result = llama->queue_results.recv(task_id);
task_result result = llama->queue_results.recv(task_id);
result_json =
result.data.dump(-1, ' ', false, json::error_handler_t::replace);
result.result_json.dump(-1, ' ', false, json::error_handler_t::replace);
resp->id = result.id;
resp->stop = result.stop;
resp->error = result.error;
@@ -363,10 +363,10 @@ void llama_server_embedding(const char *json_req, char **json_resp,
}
const int task_id = llama->queue_tasks.get_new_id();
llama->queue_results.add_waiting_task_id(task_id);
llama->request_completion(task_id, -1, {{"prompt", prompt}, {"n_predict", 0}}, false, true);
llama->request_completion(task_id, {{"prompt", prompt}, {"n_predict", 0}}, false, true, -1);
atomicRecv ar(recv_counter);
server_task_result result = llama->queue_results.recv(task_id);
std::string result_json = result.data.dump();
task_result result = llama->queue_results.recv(task_id);
std::string result_json = result.result_json.dump();
const std::string::size_type size = result_json.size() + 1;
*json_resp = new char[size];
snprintf(*json_resp, size, "%s", result_json.c_str());