CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-31 08:51:55 +00:00
Code Issues Packages Projects Releases Wiki Activity

server : simplify logic for empty prompts (#5953)

Browse Source
This commit is contained in:
Georgi Gerganov
2024-03-09 12:34:18 +02:00
committed by GitHub
parent 950ba1ab84
commit 9674aaf35c
1 changed files with 15 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files

28
examples/server/server.cpp
View File

@@ -1704,19 +1704,6 @@ struct server_context {
// next, batch any pending prompts without exceeding n_batch // next, batch any pending prompts without exceeding n_batch
if (params.cont_batching || batch.n_tokens == 0) { if (params.cont_batching || batch.n_tokens == 0) {
for (auto & slot : slots) { for (auto & slot : slots) {
const bool has_prompt = slot.prompt.is_array() || (slot.prompt.is_string() && !slot.prompt.get<std::string>().empty());
// empty prompt passed -> release the slot and send empty response
// note: infill mode allows empty prompt
if (slot.state == SLOT_STATE_IDLE && slot.command == SLOT_COMMAND_LOAD_PROMPT && !has_prompt && !slot.infill) {
slot.state = SLOT_STATE_PROCESSING;
slot.command = SLOT_COMMAND_NONE;
slot.release();
slot.print_timings();
send_final_response(slot);
continue;
}
// this slot still has a prompt to be processed // this slot still has a prompt to be processed
if (slot.state == SLOT_STATE_IDLE && slot.command == SLOT_COMMAND_LOAD_PROMPT) { if (slot.state == SLOT_STATE_IDLE && slot.command == SLOT_COMMAND_LOAD_PROMPT) {
auto & prompt_tokens = slot.prompt_tokens; auto & prompt_tokens = slot.prompt_tokens;
@@ -1768,6 +1755,21 @@ struct server_context {
{"prompt_tokens", tokens_to_str(ctx, prompt_tokens.cbegin(), prompt_tokens.cend())}, {"prompt_tokens", tokens_to_str(ctx, prompt_tokens.cbegin(), prompt_tokens.cend())},
}); });
// empty prompt passed -> release the slot and send empty response
if (prompt_tokens.empty()) {
LOG_INFO("empty prompt - releasing slot", {
{"id_slot", slot.id},
{"id_task", slot.id_task}
});
slot.state = SLOT_STATE_PROCESSING;
slot.command = SLOT_COMMAND_NONE;
slot.release();
slot.print_timings();
send_final_response(slot);
continue;
}
if (slot.embedding) { if (slot.embedding) {
// this prompt is too large to process - discard it // this prompt is too large to process - discard it
if (slot.n_prompt_tokens > n_batch) { if (slot.n_prompt_tokens > n_batch) {