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

Merge branch 'master' into concedo_experimental

Browse source 
# Conflicts:
#	Makefile
#	tests/test-grad0.cpp
#	tests/test-quantize-perf.cpp
This commit is contained in:
Concedo 2023-12-13 14:49:03 +08:00
commit c2c238b4f3
20 changed files with 44 additions and 38 deletions
Download patch file Download diff file Expand all files Collapse all files

14
llama.cpp
View file

@ -2788,7 +2788,7 @@ static void llm_load_vocab(
// The assumption is, since special tokens aren't meant to be exposed to end user, they are designed
// to be unmatchable by the tokenizer, therefore tokens from the vocab, which are unmatchable by the tokenizer
// are special tokens.
// From testing, this appears to corelate 1:1 with special tokens.
// From testing, this appears to correlate 1:1 with special tokens.
//
// Counting special tokens and verifying in only one direction
@ -5876,7 +5876,7 @@ static int llama_decode_internal(
const int64_t n_embd = hparams.n_embd;
const int64_t n_vocab = hparams.n_vocab;
// helpers for smoother batch API transistion
// helpers for smoother batch API transition
// after deprecating the llama_eval calls, these will be removed
std::vector<llama_pos> pos;
@ -6876,12 +6876,12 @@ static void tokenizer_st_partition(const llama_vocab & vocab, std::forward_list<
// loop over the text
while (true) {
// find the first occurence of a given special token in this fragment
// find the first occurrence of a given special token in this fragment
// passing offset argument only limit the "search area" but match coordinates
// are still relative to the source full raw_text
auto match = raw_text->find(special_token, raw_text_base_offset);
// no occurences found, stop processing this fragment for a given special token
// no occurrences found, stop processing this fragment for a given special token
if (match == std::string::npos) break;
// check if match is within bounds of offset <-> length
@ -7766,7 +7766,7 @@ void llama_sample_grammar(struct llama_context * ctx, llama_token_data_array * c
for (size_t i = 0; i < candidates->size; ++i) {
const llama_token id = candidates->data[i].id;
const std::string & piece = ctx->model.vocab.id_to_token[id].text;
const std::string piece = llama_token_to_piece(ctx, id);
if (id == eos) {
if (!allow_eos) {
candidates->data[i].logit = -INFINITY;
@ -7978,7 +7978,7 @@ void llama_grammar_accept_token(struct llama_context * ctx, struct llama_grammar
GGML_ASSERT(false);
}
const std::string & piece = ctx->model.vocab.id_to_token[token].text;
const std::string piece = llama_token_to_piece(ctx, token);
// Note terminating 0 in decoded string
const auto decoded = decode_utf8(piece, grammar->partial_utf8);
@ -8092,7 +8092,7 @@ struct llama_beam_search_data {
}
// Min-heaps are used to efficiently collect the top-k elements (k=n_beams).
// The repetative patterns below reflect the 2 stages of heaps:
// The repetitive patterns below reflect the 2 stages of heaps:
// * Gather elements until the vector is full, then call std::make_heap() on it.
// * If the heap is full and a new element is found that should be included, pop the
// least element to the back(), replace it with the new, then push it into the heap.