matt/ollama37
1
0
Fork 0
mirror of https://github.com/dogkeeper886/ollama37.git synced 2025-12-12 08:47:01 +00:00
Code Issues Packages Projects Releases Wiki Activity

Bump llama.cpp to b2081

Browse Source
This commit is contained in:
Daniel Hiltgen
2024-02-06 12:06:43 -08:00
parent 59ec837ef6
commit de76b95dd4
3 changed files with 17 additions and 22 deletions
Download Patch File Download Diff File Expand all files Collapse all files

27
llm/patches/02-shutdown.diff
View File

@@ -37,26 +37,18 @@ index 11dd82c3..311495a8 100644
llama_backend_free();
diff --git a/examples/server/utils.hpp b/examples/server/utils.hpp
index 70cce072..2acb1eab 100644
index 70cce072..9124869a 100644
--- a/examples/server/utils.hpp
+++ b/examples/server/utils.hpp
@@ -6,6 +6,7 @@
#include <mutex>
#include <condition_variable>
#include <unordered_map>
+#include <atomic>
#include "json.hpp"
@@ -190,6 +191,7 @@ inline std::string format_chatml(std::vector<json> messages)
@@ -190,6 +190,7 @@ inline std::string format_chatml(std::vector<json> messages)
struct llama_server_queue {
int id = 0;
std::mutex mutex_tasks;
+ std::atomic<bool> running;
+ bool running;
// queues
std::vector<task_server> queue_tasks;
std::vector<task_server> queue_tasks_deferred;
@@ -248,9 +250,15 @@ struct llama_server_queue {
@@ -248,9 +249,18 @@ struct llama_server_queue {
queue_tasks_deferred.clear();
}
@@ -64,7 +56,10 @@ index 70cce072..2acb1eab 100644
- [[noreturn]]
+ // end the start_loop routine
+ void terminate() {
+ running = false;
+ {
+ std::unique_lock<std::mutex> lock(mutex_tasks);
+ running = false;
+ }
+ condition_tasks.notify_all();
+ }
+
@@ -74,17 +69,17 @@ index 70cce072..2acb1eab 100644
while (true) {
// new task arrived
LOG_VERBOSE("have new task", {});
@@ -294,8 +302,12 @@ struct llama_server_queue {
@@ -294,8 +304,12 @@ struct llama_server_queue {
{
std::unique_lock<std::mutex> lock(mutex_tasks);
if (queue_tasks.empty()) {
+ if (!running.load()) {
+ if (!running) {
+ LOG_VERBOSE("ending start_loop", {});
+ return;
+ }
condition_tasks.wait(lock, [&]{
- return !queue_tasks.empty();
+ return (!queue_tasks.empty() || !running.load());
+ return (!queue_tasks.empty() || !running);
});
}
}