thek0tyara/llama-cpp-turboquant
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 4

server : use different seeds for child completions (#18700)

Browse source 
* server : use different seeds for child completions

* cont : handle default seed

* cont : note
This commit is contained in:
Georgi Gerganov 2026-01-09 09:33:50 +02:00 • committed by GitHub
parent 8ece3836b4
commit f5f8812f7c
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
4 changed files with 12 additions and 8 deletions
Download patch file Download diff file Expand all files Collapse all files

13
tools/server/server-context.cpp
View file

@ -4,7 +4,6 @@
#include "server-task.h"
#include "server-queue.h"
#include "arg.h"
#include "common.h"
#include "llama.h"
#include "log.h"
@ -16,7 +15,6 @@
#include <cstddef>
#include <cinttypes>
#include <memory>
#include <unordered_set>
#include <filesystem>
// fix problem with std::min and std::max
@ -2927,9 +2925,14 @@ std::unique_ptr<server_res_generator> server_routes::handle_completions_impl(
if (task.params.n_cmpl > 1) {
task.n_children = task.params.n_cmpl - 1;
for (size_t j = 0; j < task.n_children; j++) {
server_task child = task.create_child(
task.id,
rd.get_new_id());
server_task child = task.create_child(task.id, rd.get_new_id());
// use different sampling seed for each child
// note: https://github.com/ggml-org/llama.cpp/pull/18700#discussion_r2675115723
if (child.params.sampling.seed != LLAMA_DEFAULT_SEED) {
child.params.sampling.seed += j + 1;
}
tasks.push_back(std::move(child));
}
}