Skip to content

server : do not default to multiple slots with speculative decoding #17017

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

Merged
merged 2 commits into from
Nov 5, 2025
Merged

server : do not default to multiple slots with speculative decoding #17017

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
caa4ca7
server : do not default to multiple slots with speculative decoding
ggerganov Nov 5, 2025
3ce702f
cont : fix
ggerganov Nov 5, 2025
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
4 changes: 4 additions & 0 deletions common/common.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -507,6 +507,10 @@ struct common_params {
// return false from callback to abort model loading or true to continue
llama_progress_callback load_progress_callback = NULL;
void * load_progress_callback_user_data = NULL;

bool has_speculative() const {
return !speculative.model.path.empty() || !speculative.model.hf_repo.empty();
}
};

// call once at the start of a program if it uses libcommon
Expand Down
14 changes: 10 additions & 4 deletions tools/server/server.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2400,7 +2400,7 @@ struct server_context {

add_bos_token = llama_vocab_get_add_bos(vocab);

if (!params_base.speculative.model.path.empty() || !params_base.speculative.model.hf_repo.empty()) {
if (params_base.has_speculative()) {
SRV_INF("loading draft model '%s'\n", params_base.speculative.model.path.c_str());

auto params_dft = params_base;
Expand Down Expand Up @@ -2476,7 +2476,7 @@ struct server_context {
SRV_WRN("%s\n", "cache_reuse is not supported by multimodal, it will be disabled");
}

if (!params_base.speculative.model.path.empty()) {
if (params_base.has_speculative()) {
SRV_ERR("%s\n", "err: speculative decode is not supported by multimodal");
return false;
}
Expand Down Expand Up @@ -2520,6 +2520,7 @@ struct server_context {
if (model_dft) {
slot.batch_spec = llama_batch_init(params_base.speculative.n_max + 1, 0, 1);

// TODO: rework speculative decoding [TAG_SERVER_SPEC_REWORK]
slot.ctx_dft = llama_init_from_model(model_dft, cparams_dft);
if (slot.ctx_dft == nullptr) {
SRV_ERR("%s", "failed to create draft context\n");
Expand Down Expand Up @@ -2825,6 +2826,7 @@ struct server_context {
}

// initialize draft batch
// TODO: rework speculative decoding [TAG_SERVER_SPEC_REWORK]
if (slot.ctx_dft) {
llama_batch_free(slot.batch_spec);

Expand Down Expand Up @@ -4291,6 +4293,8 @@ struct server_context {
}

// do speculative decoding
// TODO: rework to have a single draft llama_context shared across all slots [TAG_SERVER_SPEC_REWORK]
// perform the speculative drafting for all sequences at the same time in a single batch
for (auto & slot : slots) {
if (!slot.is_processing() || !slot.can_speculate()) {
continue;
Expand Down Expand Up @@ -4445,8 +4449,10 @@ int main(int argc, char ** argv) {

// TODO: should we have a separate n_parallel parameter for the server?
// https://github.com/ggml-org/llama.cpp/pull/16736#discussion_r2483763177
if (params.n_parallel == 1 && params.kv_unified == false) {
LOG_WRN("%s: setting n_parallel = 4 and kv_unified = true\n", __func__);
// TODO: this is a common configuration that is suitable for most local use cases
// however, overriding the parameters is a bit confusing - figure out something more intuitive
if (params.n_parallel == 1 && params.kv_unified == false && !params.has_speculative()) {
LOG_WRN("%s: setting n_parallel = 4 and kv_unified = true (add -kvu to disable this)\n", __func__);

params.n_parallel = 4;
params.kv_unified = true;
Expand Down
Loading