default penalize_nl to false + format

This commit is contained in:
anon 2023-05-31 11:56:12 -03:00
parent 40e13805d9
commit d58e48663d

View file

@ -507,210 +507,151 @@ bool server_params_parse(int argc, char **argv, server_params &sparams, gpt_para
return true; return true;
} }
bool parse_options_completion(json body, llama_server_context& llama, Response &res) { bool parse_options_completion(json body, llama_server_context& llama, Response &res)
{
gpt_params default_params; gpt_params default_params;
if (!body["stream"].is_null()) if (!body["stream"].is_null()) {
{ llama.stream = body["stream"].get<bool>();
llama.stream = body["stream"].get<bool>(); } else {
llama.stream = false;
} }
else if (!body["n_predict"].is_null()) {
{ llama.params.n_predict = body["n_predict"].get<int>();
llama.stream = false; } else {
llama.params.n_predict = default_params.n_predict;
} }
if (!body["n_predict"].is_null()) if (!body["top_k"].is_null()) {
{ llama.params.top_k = body["top_k"].get<int>();
llama.params.n_predict = body["n_predict"].get<int>(); } else {
llama.params.top_k = default_params.top_k;
} }
else if (!body["top_p"].is_null()) {
{ llama.params.top_p = body["top_p"].get<float>();
llama.params.n_predict = default_params.n_predict; } else {
llama.params.top_p = default_params.top_p;
} }
if (!body["top_k"].is_null()) if (!body["tfs_z"].is_null()) {
{ llama.params.tfs_z = body["tfs_z"].get<float>();
llama.params.top_k = body["top_k"].get<int>(); } else {
llama.params.tfs_z = default_params.tfs_z;
} }
else if (!body["typical_p"].is_null()) {
{ llama.params.typical_p = body["typical_p"].get<float>();
llama.params.top_k = default_params.top_k; } else {
llama.params.typical_p = default_params.typical_p;
} }
if (!body["top_p"].is_null()) if (!body["repeat_last_n"].is_null()) {
{ llama.params.repeat_last_n = body["repeat_last_n"].get<int>();
llama.params.top_p = body["top_p"].get<float>(); } else {
llama.params.repeat_last_n = default_params.repeat_last_n;
} }
else if (!body["temperature"].is_null()) {
{ llama.params.temp = body["temperature"].get<float>();
llama.params.top_p = default_params.top_p; } else {
llama.params.temp = default_params.temp;
} }
if (!body["tfs_z"].is_null()) if (!body["repeat_penalty"].is_null()) {
{ llama.params.repeat_penalty = body["repeat_penalty"].get<float>();
llama.params.tfs_z = body["tfs_z"].get<float>(); } else {
llama.params.repeat_penalty = default_params.repeat_penalty;
} }
else if (!body["presence_penalty"].is_null()) {
{ llama.params.presence_penalty = body["presence_penalty"].get<float>();
llama.params.tfs_z = default_params.tfs_z; } else {
llama.params.presence_penalty = default_params.presence_penalty;
} }
if (!body["typical_p"].is_null()) if (!body["frequency_penalty"].is_null()) {
{ llama.params.frequency_penalty = body["frequency_penalty"].get<float>();
llama.params.typical_p = body["typical_p"].get<float>(); } else {
llama.params.frequency_penalty = default_params.frequency_penalty;
} }
else if (!body["mirostat"].is_null()) {
{ llama.params.mirostat = body["mirostat"].get<float>();
llama.params.typical_p = default_params.typical_p; } else {
llama.params.mirostat = default_params.mirostat;
} }
if (!body["repeat_last_n"].is_null()) if (!body["mirostat_tau"].is_null()) {
{ llama.params.mirostat_tau = body["mirostat_tau"].get<float>();
llama.params.repeat_last_n = body["repeat_last_n"].get<int>(); } else {
llama.params.mirostat_tau = default_params.mirostat_tau;
} }
else if (!body["mirostat_eta"].is_null()) {
{ llama.params.mirostat_eta = body["mirostat_eta"].get<float>();
llama.params.repeat_last_n = default_params.repeat_last_n; } else {
llama.params.mirostat_eta = default_params.mirostat_eta;
} }
if (!body["temperature"].is_null()) if (!body["penalize_nl"].is_null()) {
{ llama.params.penalize_nl = body["penalize_nl"].get<float>();
llama.params.temp = body["temperature"].get<float>(); } else {
llama.params.penalize_nl = false;
} }
else if (!body["n_keep"].is_null()) {
{ llama.params.n_keep = body["n_keep"].get<int>();
llama.params.temp = default_params.temp; } else {
llama.params.n_keep = default_params.n_keep;
} }
if (!body["repeat_penalty"].is_null()) if (!body["seed"].is_null()) {
{
llama.params.repeat_penalty = body["repeat_penalty"].get<float>();
}
else
{
llama.params.repeat_penalty = default_params.repeat_penalty;
}
if (!body["presence_penalty"].is_null())
{
llama.params.presence_penalty = body["presence_penalty"].get<float>();
}
else
{
llama.params.presence_penalty = default_params.presence_penalty;
}
if (!body["frequency_penalty"].is_null())
{
llama.params.frequency_penalty = body["frequency_penalty"].get<float>();
}
else
{
llama.params.frequency_penalty = default_params.frequency_penalty;
}
if (!body["mirostat"].is_null())
{
llama.params.mirostat = body["mirostat"].get<float>();
}
else
{
llama.params.mirostat = default_params.mirostat;
}
if (!body["mirostat_tau"].is_null())
{
llama.params.mirostat_tau = body["mirostat_tau"].get<float>();
}
else
{
llama.params.mirostat_tau = default_params.mirostat_tau;
}
if (!body["mirostat_eta"].is_null())
{
llama.params.mirostat_eta = body["mirostat_eta"].get<float>();
}
else
{
llama.params.mirostat_eta = default_params.mirostat_eta;
}
if (!body["penalize_nl"].is_null())
{
llama.params.penalize_nl = body["penalize_nl"].get<float>();
}
else
{
llama.params.penalize_nl = default_params.penalize_nl;
}
if (!body["n_keep"].is_null())
{
llama.params.n_keep = body["n_keep"].get<int>();
}
else
{
llama.params.n_keep = default_params.n_keep;
}
if (!body["seed"].is_null())
{
llama.params.seed = body["seed"].get<int>(); llama.params.seed = body["seed"].get<int>();
} } else {
else
{
llama.params.seed = time(NULL); llama.params.seed = time(NULL);
} }
if (!body["ignore_eos"].is_null() && body["ignore_eos"].get<bool>()) if (!body["ignore_eos"].is_null() && body["ignore_eos"].get<bool>()) {
{ llama.params.logit_bias[llama_token_eos()] = -INFINITY;
llama.params.logit_bias[llama_token_eos()] = -INFINITY; } else {
llama.params.logit_bias.erase(llama_token_eos());
} }
else if (!body["prompt"].is_null()) {
{
llama.params.logit_bias.erase(llama_token_eos());
}
if (!body["prompt"].is_null())
{
llama.params.prompt = body["prompt"].get<std::string>(); llama.params.prompt = body["prompt"].get<std::string>();
} } else {
else json data = {{"status", "error"}, {"reason", "You need to pass the prompt"}};
{
json data = {
{"status", "error"},
{"reason", "You need to pass the prompt"}};
res.set_content(data.dump(llama.json_indent), "application/json"); res.set_content(data.dump(llama.json_indent), "application/json");
res.status = 400; res.status = 400;
return false; return false;
} }
if (!body["stop"].is_null()) if (!body["stop"].is_null()) {
{
llama.params.antiprompt = body["stop"].get<std::vector<std::string>>(); llama.params.antiprompt = body["stop"].get<std::vector<std::string>>();
} } else {
else llama.params.antiprompt.clear();
{
llama.params.antiprompt.clear();
} }
if (llama.verbose) { if (llama.verbose) {
std::string tmp_stop = std::string tmp_stop =
std::accumulate(llama.params.antiprompt.begin(), llama.params.antiprompt.end(), std::accumulate(llama.params.antiprompt.begin(), llama.params.antiprompt.end(),
std::string{}, [](std::string a, std::string b) { std::string{}, [](std::string a, std::string b) {
return a + (a != "" ? ", \"" : "\"") + b + "\""; return a + (a != "" ? ", \"" : "\"") + b + "\"";
}); });
fprintf(stderr, fprintf(stderr,
"-------------------------\n" "-------------------------\n"
"/completion parameters: {\n" "/completion parameters: {\n"
" stream: %d,\n" " stream: %d,\n"
" frequency_penalty: %f,\n" " ignore_eos: %d,\n"
" mirostat: %d,\n" " frequency_penalty: %f,\n"
" mirostat_eta: %f,\n" " mirostat: %d,\n"
" mirostat_tau: %f,\n" " mirostat_eta: %f,\n"
" n_keep: %d,\n" " mirostat_tau: %f,\n"
" n_predict: %d,\n" " n_keep: %d,\n"
" penalize_nl: %d,\n" " n_predict: %d,\n"
" presence_penalty: %f,\n" " penalize_nl: %d,\n"
" repeat_last_n: %d,\n" " presence_penalty: %f,\n"
" repeat_penalty: %f,\n" " repeat_last_n: %d,\n"
" seed: %d,\n" " repeat_penalty: %f,\n"
" stop: [%s],\n" " seed: %d,\n"
" temperature: %f,\n" " stop: [%s],\n"
" tfs_z: %f,\n" " temperature: %f,\n"
" top_k: %d,\n" " tfs_z: %f,\n"
" top_p: %f,\n" " top_k: %d,\n"
" typical_p: %f,\n" " top_p: %f,\n"
"}\nPROMPT[%s]\n", " typical_p: %f,\n"
llama.stream, llama.params.frequency_penalty, llama.params.mirostat, "}\nPROMPT[%s]\n",
llama.params.mirostat_eta, llama.params.mirostat_tau, llama.params.n_keep, llama.stream, -INFINITY == llama.params.logit_bias[llama_token_eos()],
llama.params.n_predict, llama.params.penalize_nl, llama.params.frequency_penalty, llama.params.mirostat,
llama.params.presence_penalty, llama.params.repeat_last_n, llama.params.mirostat_eta, llama.params.mirostat_tau, llama.params.n_keep,
llama.params.repeat_penalty, llama.params.seed, tmp_stop.c_str(), llama.params.n_predict, llama.params.penalize_nl,
llama.params.temp, llama.params.tfs_z, llama.params.top_k, llama.params.presence_penalty, llama.params.repeat_last_n,
llama.params.top_p, llama.params.typical_p, llama.params.prompt.c_str()); llama.params.repeat_penalty, llama.params.seed, tmp_stop.c_str(),
llama.params.temp, llama.params.tfs_z, llama.params.top_k, llama.params.top_p,
llama.params.typical_p, llama.params.prompt.c_str());
} }
return true; return true;