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

llama : refactor sampling v2 (#9294)

Browse source 
- Add `struct llama_sampler` and `struct llama_sampler_i`
- Add `llama_sampler_` API
- Add `llama_sampler_chain_` API for chaining multiple samplers
- Remove `LLAMA_API_INTERNAL`
- Add `llama_perf_` API and remove old `llama_print_timings` and `llama_reset_timings`
This commit is contained in:
Georgi Gerganov 2024-09-07 15:16:19 +03:00 • committed by GitHub
parent 947538acb8
commit df270ef745
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
48 changed files with 3497 additions and 2914 deletions
Download patch file Download diff file Expand all files Collapse all files

29
examples/simple/simple.cpp
View file

@ -55,6 +55,14 @@ int main(int argc, char ** argv) {
return 1;
}
auto sparams = llama_sampler_chain_default_params();
sparams.no_perf = false;
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;
@ -110,20 +118,9 @@ int main(int argc, char ** argv) {
while (n_cur <= n_predict) {
// sample the next token
{
auto n_vocab = llama_n_vocab(model);
auto * logits = llama_get_logits_ith(ctx, batch.n_tokens - 1);
const llama_token new_token_id = llama_sampler_sample(smpl, 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);
llama_sampler_accept(smpl, new_token_id);
// is it an end of generation?
if (llama_token_is_eog(model, new_token_id) || n_cur == n_predict) {
@ -160,12 +157,14 @@ 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_print(smpl, LLAMA_PERF_TYPE_SAMPLER_CHAIN);
llama_perf_print(ctx, LLAMA_PERF_TYPE_CONTEXT);
fprintf(stderr, "\n");
llama_batch_free(batch);
llama_sampler_free(smpl);
llama_free(ctx);
llama_free_model(model);