Mirrors/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2025-01-06 02:48:57 +01:00
Code Issues Packages Projects Releases Wiki Activity

server : some style changes

Browse Source
This commit is contained in:
Georgi Gerganov 2023-11-24 10:49:08 +02:00
parent e1516709f2
commit f25308be5c
No known key found for this signature in database
GPG Key ID: 449E073F9DC10735
1 changed files with 53 additions and 57 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

28
examples/server/server.cpp
View File

@ -385,7 +385,7 @@ struct llama_client_slot
bool stopped_limit = false; bool stopped_limit = false;
bool oaicompat = false; bool oaicompat = false;
std::string oaicompat_model = ""; std::string oaicompat_model;
std::string stopping_word; std::string stopping_word;
@ -486,7 +486,7 @@ struct llama_client_slot
}; };
} }
void print_timings() { void print_timings() const {
LOG_TEE("\n"); LOG_TEE("\n");
LOG_TEE("%s: prompt eval time = %10.2f ms / %5d tokens (%8.2f ms per token, %8.2f tokens per second)\n", LOG_TEE("%s: prompt eval time = %10.2f ms / %5d tokens (%8.2f ms per token, %8.2f tokens per second)\n",
__func__, t_prompt_processing, num_prompt_tokens_processed, t_prompt_processing / num_prompt_tokens_processed, 1e3 / t_prompt_processing * num_prompt_tokens_processed); __func__, t_prompt_processing, num_prompt_tokens_processed, t_prompt_processing / num_prompt_tokens_processed, 1e3 / t_prompt_processing * num_prompt_tokens_processed);
@ -1284,7 +1284,7 @@ struct llama_server_context
std::lock_guard<std::mutex> lock(mutex_tasks); std::lock_guard<std::mutex> lock(mutex_tasks);
task_server task; task_server task;
task.id = id_gen++; task.id = id_gen++;
task.data = data; task.data = std::move(data);
task.infill_mode = infill; task.infill_mode = infill;
task.embedding_mode = embedding; task.embedding_mode = embedding;
task.type = COMPLETION_TASK; task.type = COMPLETION_TASK;
@ -2254,14 +2254,14 @@ json oaicompat_completion_params_parse(
// Map OpenAI parameters to llama.cpp parameters // Map OpenAI parameters to llama.cpp parameters
llama_params["prompt"] = format_chatml(body["messages"]); // OpenAI 'messages' to llama.cpp 'prompt' llama_params["prompt"] = format_chatml(body["messages"]); // OpenAI 'messages' to llama.cpp 'prompt'
llama_params["temperature"] = json_value(body, "temperature", 0.8); llama_params["temperature"] = json_value(body, "temperature", 0.8);
llama_params["top_k"] = json_value(body, "max_tokens", 40); llama_params["top_k"] = json_value(body, "top_k", 40);
llama_params["top_p"] = json_value(body, "top_p", 0.95); llama_params["top_p"] = json_value(body, "top_p", 0.95);
llama_params["n_predict"] = json_value(body, "max_tokens", -1); llama_params["n_predict"] = json_value(body, "max_tokens", -1);
llama_params["logit_bias"] = json_value(body, "logit_bias",json::object()); llama_params["logit_bias"] = json_value(body, "logit_bias",json::object());
llama_params["frequency_penalty"] = json_value(body, "frequency_penalty", 0.0); llama_params["frequency_penalty"] = json_value(body, "frequency_penalty", 0.0);
llama_params["presence_penalty"] = json_value(body, "presence_penalty", 0.0); llama_params["presence_penalty"] = json_value(body, "presence_penalty", 0.0);
llama_params["seed"] = json_value(body, "seed", 0); llama_params["seed"] = json_value(body, "seed", 0);
llama_params["stream"] =json_value(body, "stream", false); llama_params["stream"] = json_value(body, "stream", false);
llama_params["mirostat"] = json_value(body, "mirostat", false); llama_params["mirostat"] = json_value(body, "mirostat", false);
llama_params["mirostat_tau"] = json_value(body, "mirostat_tau", 0.0); llama_params["mirostat_tau"] = json_value(body, "mirostat_tau", 0.0);
llama_params["mirostat_eta"] = json_value(body, "mirostat_eta", 0.0); llama_params["mirostat_eta"] = json_value(body, "mirostat_eta", 0.0);
@ -2272,9 +2272,7 @@ json oaicompat_completion_params_parse(
llama_params["tfs_z"] = json_value(body, "tfs_z", 0.0); llama_params["tfs_z"] = json_value(body, "tfs_z", 0.0);
if (llama_params.count("grammar") != 0) { if (llama_params.count("grammar") != 0) {
llama_params["grammar"] = json_value( llama_params["grammar"] = json_value(body, "grammar", json::object());
body, "grammar",
json::object());
} }
// Handle 'stop' field // Handle 'stop' field
@ -2294,8 +2292,7 @@ json oaicompat_completion_params_parse(
return llama_params; return llama_params;
} }
static json format_final_response_oaicompat(json request, task_result response, static json format_final_response_oaicompat(const json &request, const task_result &response, bool streaming = false)
bool streaming = false)
{ {
json result = response.result_json; json result = response.result_json;
@ -2345,7 +2342,7 @@ static json format_final_response_oaicompat(json request, task_result response,
} }
// return value is vector as there is one case where we might need to generate two responses // return value is vector as there is one case where we might need to generate two responses
static std::vector<json> format_partial_response_oaicompat(task_result response) { static std::vector<json> format_partial_response_oaicompat(const task_result &response) {
json result = response.result_json; json result = response.result_json;
if (!result.contains("model") || !result.contains("oaicompat_token_ctr")) { if (!result.contains("model") || !result.contains("oaicompat_token_ctr")) {
@ -2353,15 +2350,14 @@ static std::vector<json> format_partial_response_oaicompat(task_result response)
} }
bool first = json_value(result, "oaicompat_token_ctr", 0) == 0; bool first = json_value(result, "oaicompat_token_ctr", 0) == 0;
std::string modelname = std::string modelname = json_value(result, "model", std::string(DEFAULT_OAICOMPAT_MODEL));
json_value(result, "model", std::string(DEFAULT_OAICOMPAT_MODEL));
bool stopped_word = json_value(result, "stopped_word", false); bool stopped_word = json_value(result, "stopped_word", false);
bool stopped_eos = json_value(result, "stopped_eos", false); bool stopped_eos = json_value(result, "stopped_eos", false);
bool stopped_limit = json_value(result, "stopped_limit", false); bool stopped_limit = json_value(result, "stopped_limit", false);
std::string content = json_value(result, "content", std::string("")); std::string content = json_value(result, "content", std::string(""));
std::string finish_reason = ""; std::string finish_reason;
if (stopped_word || stopped_eos) { if (stopped_word || stopped_eos) {
finish_reason = "stop"; finish_reason = "stop";
} }
@ -2655,8 +2651,8 @@ int main(int argc, char **argv)
}); });
svr.Post("/v1/chat/completions", [&llama](const httplib::Request &req, // TODO: add mount point without "/v1" prefix -- how?
httplib::Response &res) svr.Post("/v1/chat/completions", [&llama](const httplib::Request &req, httplib::Response &res)
{ {
json data = oaicompat_completion_params_parse(json::parse(req.body)); json data = oaicompat_completion_params_parse(json::parse(req.body));