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

llama : refactor samplers internal implementation (#9370)

Browse source
This commit is contained in:
slaren 2024-09-08 15:52:07 +02:00 • committed by GitHub
parent 2a358fb0c4
commit 19f4a7b296
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
4 changed files with 881 additions and 725 deletions
Download patch file Download diff file Expand all files Collapse all files

4
src/llama-impl.h
View file

@ -101,6 +101,10 @@ struct ring_buffer {
}
void push_back(const T & value) {
if (capacity == 0) {
throw std::runtime_error("ring buffer: capacity is zero");
}
if (sz == capacity) {
// advance the start when buffer is full
first = (first + 1) % capacity;

1582
src/llama-sampling.cpp
View file

File diff suppressed because it is too large Load diff

10
src/llama-sampling.h
View file

@ -23,16 +23,6 @@ struct llama_sampler_chain {
mutable int32_t n_sample;
};
using llama_token_cnt = std::unordered_map<llama_token, int>;
// TODO: tmp exposed until test-sampling is fixed
void llama_sampler_penalties_impl(
llama_token_data_array * cur_p,
const llama_token_cnt & token_count,
float penalty_repeat,
float penalty_freq,
float penalty_present);
struct llama_sampler * llama_sampler_init_grammar_impl(
const struct llama_vocab & vocab,
const char * grammar_str,