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

expose timing info in web api

Browse source
This commit is contained in:
Concedo 2023-07-11 18:56:06 +08:00
parent 7222877069
commit 1d1111e10f
4 changed files with 17 additions and 16 deletions
Download patch file Download diff file Expand all files Collapse all files

8
expose.cpp
View file

@ -220,12 +220,12 @@ extern "C"
return generation_finished;
}
float get_prompt_eval_time() {
return prompt_eval_time;
float get_last_eval_time() {
return last_eval_time;
}
float get_prompt_process_time() {
return prompt_process_time;
float get_last_process_time() {
return last_process_time;
}
const char* get_pending_output() {

4
expose.h
View file

@ -72,5 +72,5 @@ extern std::string lora_filename;
extern std::string lora_base;
extern std::vector<std::string> generated_tokens;
extern bool generation_finished;
extern float prompt_eval_time;
extern float prompt_process_time;
extern float last_eval_time;
extern float last_process_time;

10
gpttype_adapter.cpp
View file

@ -33,8 +33,8 @@ std::string executable_path = "";
std::string lora_filename = "";
std::string lora_base = "";
bool generation_finished;
float prompt_process_time;
float prompt_eval_time;
float last_process_time = 0;
float last_eval_time = 0;
std::vector<std::string> generated_tokens;
//return val: 0=fail, 1=(original ggml, alpaca), 2=(ggmf), 3=(ggjt)
@ -869,8 +869,6 @@ generation_outputs gpttype_generate(const generation_inputs inputs, generation_o
bool stream_sse = inputs.stream_sse;
generation_finished = false; // Set current generation status
prompt_eval_time = 0;
prompt_process_time = 0;
generated_tokens.clear(); // New Generation, new tokens
if (params.repeat_last_n < 1)
@ -1449,8 +1447,8 @@ generation_outputs gpttype_generate(const generation_inputs inputs, generation_o
fflush(stdout);
output.status = 1;
generation_finished = true;
prompt_eval_time = pt2;
prompt_process_time = pt1;
last_eval_time = pt2;
last_process_time = pt1;
snprintf(output.text, sizeof(output.text), "%s", concat_output.c_str());
return output;

11
koboldcpp.py
View file

@ -161,8 +161,8 @@ def init_library():
handle.new_token.argtypes = [ctypes.c_int]
handle.get_stream_count.restype = ctypes.c_int
handle.has_finished.restype = ctypes.c_bool
handle.get_prompt_eval_time.restype = ctypes.c_float
handle.get_prompt_process_time.restype = ctypes.c_float
handle.get_last_eval_time.restype = ctypes.c_float
handle.get_last_process_time.restype = ctypes.c_float
handle.abort_generate.restype = ctypes.c_bool
handle.get_pending_output.restype = ctypes.c_char_p
@ -455,6 +455,11 @@ class ServerRequestHandler(http.server.SimpleHTTPRequestHandler):
elif self.path.endswith(('/api/extra/version')):
response_body = (json.dumps({"result":"KoboldCpp","version":KcppVersion}).encode())
elif self.path.endswith(('/api/extra/perf')):
lastp = handle.get_last_process_time()
laste = handle.get_last_eval_time()
response_body = (json.dumps({"last_process":lastp,"last_eval":laste}).encode())
if response_body is None:
self.send_response(404)
self.end_headers()
@ -532,8 +537,6 @@ class ServerRequestHandler(http.server.SimpleHTTPRequestHandler):
newprompt = fullprompt
gen = asyncio.run(self.handle_request(genparams, newprompt, basic_api_flag, kai_sse_stream_flag))
gen['prompt_process_time'] = handle.get_prompt_process_time()
gen['prompt_eval_time'] = handle.get_prompt_eval_time()
try:
self.send_response(200)
self.end_headers()