vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

avoid to get prompt in infill mode and embedding mode

Browse source
This commit is contained in:
wudexiang 2024-05-15 01:54:13 +08:00
parent 9f773486ab
commit 2992479a42
1 changed files with 14 additions and 11 deletions
Download patch file Download diff file Expand all files Collapse all files

25
examples/server/server.cpp
View file

@ -901,20 +901,23 @@ struct server_context {
slot.params.input_suffix = json_value(data, "input_suffix", default_params.input_suffix);
// get prompt
{
const auto & prompt = data.find("prompt");
if (prompt == data.end()) {
send_error(task, "Either \"prompt\" or \"messages\" must be provided", ERROR_TYPE_INVALID_REQUEST);
return false;
} else {
slot.prompt = *prompt;
}
if (slot.prompt.is_array() && slot.prompt.size() == 0) {
send_error(task, "\"prompt\" cannot be an empty array", ERROR_TYPE_INVALID_REQUEST);
return false;
if (!(task.infill || task.embedding)) {
{
const auto & prompt = data.find("prompt");
if (prompt == data.end()) {
send_error(task, "Either \"prompt\" or \"messages\" must be provided", ERROR_TYPE_INVALID_REQUEST);
return false;
} else {
slot.prompt = *prompt;
}
if (slot.prompt.is_array() && slot.prompt.size() == 0) {
send_error(task, "\"prompt\" cannot be an empty array", ERROR_TYPE_INVALID_REQUEST);
return false;
}
}
}
// penalize user-provided tokens
{
slot.sparams.penalty_prompt_tokens.clear();