vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2025-09-11 01:24:36 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

still refactoring, but need a checkpoint to prepare build for 1.0.7

Browse source
This commit is contained in:
Concedo 2023-04-01 08:55:14 +08:00
parent 6b86f5ea22
commit 801b178f2a
11 changed files with 452 additions and 44 deletions
Download patch file Download diff file Expand all files Collapse all files

38
llama_adapter.cpp
View file

@ -9,12 +9,11 @@
#include <time.h>
#include "./examples/main/main.cpp"
#include "extra.h"
#include "ggml.h"
#include "llama_adapter.h"
#include "model_adapter.h"
//return val: 0=fail, 1=(original ggml, alpaca), 2=(ggmf), 3=(ggjt)
static int file_format = 0;
static FileFormat file_format = FileFormat::FAIL;
static llama_context_params ctx_params;
static gpt_params params;
static int n_past = 0;
@ -25,36 +24,7 @@ static llama_context *ctx;
static std::vector<llama_token> last_n_tokens;
static std::vector<llama_token> current_context_tokens;
static clock_t bench_timer = 0;
void timer_start()
{
bench_timer = clock();
}
double timer_check()
{
double ticks = clock() - bench_timer;
double time_taken = ((double)ticks) / CLOCKS_PER_SEC;
return time_taken;
}
void print_tok_vec(std::vector<llama_token> &embd)
{
std::cout << "[";
bool first = true;
for (auto i : embd)
{
if (!first)
{
std::cout << ',';
}
first = false;
std::cout << i;
}
std::cout << "]";
}
bool llama_load_model(const load_model_inputs inputs, FileFormat file_format)
bool llama_load_model(const load_model_inputs inputs, FileFormat in_file_format)
{
printf("System Info: %s\n", llama_print_system_info());
@ -70,6 +40,8 @@ bool llama_load_model(const load_model_inputs inputs, FileFormat file_format)
ctx_params.f16_kv = inputs.f16_kv;
ctx_params.logits_all = false;
file_format = in_file_format;
if (file_format == FileFormat::GGML || file_format == FileFormat::GGHF)
{
ctx = legacy_llama_init_from_file(model.c_str(), ctx_params);