Skip to content

Fix slot selection logic in get_available_slot #1332

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
CamNoob wants to merge 3 commits into
base: main
Choose a base branch
from
+8 −9
Open

Fix slot selection logic in get_available_slot #1332

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
14e58c4
Fix slot selection logic in get_available_slot
Feb 27, 2026
cc069f8
Fix ncmoe interaction with self-extend (sm) layer
Feb 27, 2026
d97d6b1
Revert "Fix ncmoe interaction with self-extend (sm) layer"
Feb 28, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
17 changes: 8 additions & 9 deletions examples/server/server-context.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -690,9 +690,10 @@ server_slot* server_context::get_available_slot(const server_task& task) {
bool update_cache = false;

// find the slot that has at least n% prompt similarity
if (ret == nullptr && slot_prompt_similarity != 0.0f) {
if (slot_prompt_similarity != 0.0f) {
int max_lcp_len = 0;
float sim_best = 0;
float sim_best = -1.0f;
server_slot* best_slot = nullptr;

for (server_slot& slot : slots) {
// skip the slot if it is not available
Expand Down Expand Up @@ -720,13 +721,14 @@ server_slot* server_context::get_available_slot(const server_task& task) {
float sim_cur = sim.second;

// select the current slot if the criteria match
if (sim_cur > sim_best && sim_cur > slot_prompt_similarity) {
if (sim_cur > slot_prompt_similarity && sim_cur > sim_best) {
sim_best = sim_cur;
max_lcp_len = lcp_len.first;
ret = &slot;
best_slot = &slot;
}
}
if (ret != nullptr) {
if (best_slot != nullptr) {
ret = best_slot;
LOG_VERBOSE("selected slot by lcp similarity", {
{"id_slot", ret->id},
{"max_lcp_len", max_lcp_len},
Expand All @@ -749,7 +751,6 @@ server_slot* server_context::get_available_slot(const server_task& task) {
ret = &slot;
}
}

if (ret != nullptr) {
LOG_VERBOSE("selected slot by lru", {
{"id_slot", ret->id},
Expand All @@ -766,7 +767,6 @@ server_slot* server_context::get_available_slot(const server_task& task) {
if (exclude_think) {
auto temp = tokens.get_text_tokens_exclude_think(ret->ctx, ret->params.think_tokens);
server_tokens cache_exclude_think = server_tokens(temp, false);

temp = task.tokens.get_text_tokens_exclude_think(ret->ctx, ret->params.think_tokens);
server_tokens prompt_exclude_think = server_tokens(temp, false);

Expand Down Expand Up @@ -799,15 +799,13 @@ server_slot* server_context::get_available_slot(const server_task& task) {
LLAMA_LOG_INFO("updating prompt cache\n");
// copy cache tokens
copy_data_to_cached_prompt(tokens, *ret);

ret->prompt_save(*prompt_cache);
LLAMA_LOG_INFO("prompt cache save took %.2f ms\n", (ggml_time_us() - t_start) / 1000.0);
}
// has prompts saved earlier to load
if (prompt_cache && !prompt_cache->states.empty()) {
const int64_t t_start = ggml_time_us();
copy_data_to_cached_prompt(tokens, *ret);

ret->prompt_load(*prompt_cache, task.tokens);
prompt_cache->update();

Expand All @@ -821,6 +819,7 @@ server_slot* server_context::get_available_slot(const server_task& task) {
return ret;
}


bool server_context::launch_slot_with_task(server_slot& slot, server_task& task) {
slot_params defaults;
defaults.speculative = params_base.speculative;
Expand Down