CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-11-01 09:01:57 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'master' into compilade/refactor-kv-cache

Browse Source
This commit is contained in:
Francis Couture-Harpin
2024-09-14 16:08:52 -04:00
parent 4bb4b22a58 822b6322de
commit 63ac36b271
144 changed files with 11344 additions and 6693 deletions
Download Patch File Download Diff File Expand all files Collapse all files

37
examples/passkey/passkey.cpp
View File

@@ -1,3 +1,4 @@
#include "arg.h"
#include "common.h"
#include "llama.h"
@@ -6,9 +7,7 @@
#include <string>
#include <vector>
static void print_usage(int argc, char ** argv, const gpt_params & params) {
gpt_params_print_usage(argc, argv, params);
static void print_usage(int, char ** argv) {
LOG_TEE("\nexample usage:\n");
LOG_TEE("\n %s -m model.gguf --junk 250 --pos 90 --keep 32 --grp-attn-n 2 [--seed 1234]\n", argv[0]);
LOG_TEE("\n");
@@ -21,13 +20,10 @@ int main(int argc, char ** argv) {
params.n_keep = 32;
params.i_pos = -1;
if (!gpt_params_parse(argc, argv, params)) {
print_usage(argc, argv, params);
if (!gpt_params_parse(argc, argv, params, LLAMA_EXAMPLE_PASSKEY, print_usage)) {
return 1;
}
srand(params.seed == LLAMA_DEFAULT_SEED ? time(NULL) : params.seed);
int n_junk = params.n_junk;
int n_keep = params.n_keep;
int n_grp = params.grp_attn_n;
@@ -80,12 +76,17 @@ int main(int argc, char ** argv) {
GGML_ASSERT(ctx_params.n_batch % n_grp == 0 && "n_batch must be divisible by n_grp");
llama_context * ctx = llama_new_context_with_model(model, ctx_params);
if (ctx == NULL) {
fprintf(stderr , "%s: error: failed to create the llama_context\n" , __func__);
return 1;
}
auto sparams = llama_sampler_chain_default_params();
llama_sampler * smpl = llama_sampler_chain_init(sparams);
llama_sampler_chain_add(smpl, llama_sampler_init_greedy());
// tokenize the prompt
std::vector<llama_token> tokens_list;
tokens_list = ::llama_tokenize(ctx, params.prompt, true);
@@ -217,20 +218,7 @@ int main(int argc, char ** argv) {
while (n_cur <= n_len) {
// sample the next token
{
auto n_vocab = llama_n_vocab(model);
auto * logits = llama_get_logits_ith(ctx, batch.n_tokens - 1);
std::vector<llama_token_data> candidates;
candidates.reserve(n_vocab);
for (llama_token token_id = 0; token_id < n_vocab; token_id++) {
candidates.emplace_back(llama_token_data{ token_id, logits[token_id], 0.0f });
}
llama_token_data_array candidates_p = { candidates.data(), candidates.size(), false };
// sample the most likely token
const llama_token new_token_id = llama_sample_token_greedy(ctx, &candidates_p);
const llama_token new_token_id = llama_sampler_sample(smpl, ctx, batch.n_tokens - 1);
// is it an end of generation?
if (llama_token_is_eog(model, new_token_id) || n_cur == n_len) {
@@ -267,10 +255,13 @@ int main(int argc, char ** argv) {
LOG_TEE("%s: decoded %d tokens in %.2f s, speed: %.2f t/s\n",
__func__, n_decode, (t_main_end - t_main_start) / 1000000.0f, n_decode / ((t_main_end - t_main_start) / 1000000.0f));
llama_print_timings(ctx);
LOG_TEE("\n");
llama_perf_context_print(ctx);
fprintf(stderr, "\n");
llama_sampler_free(smpl);
llama_batch_free(batch);
llama_free(ctx);