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

remove debug prints for now, they were kind of cluttered

Browse source
This commit is contained in:
Concedo 2025-06-13 16:00:23 +08:00
parent 5af9138ebe
commit 5bac0fb3d5
1 changed files with 6 additions and 13 deletions
Download patch file Download diff file Expand all files Collapse all files

19
gpttype_adapter.cpp
View file

@ -1593,7 +1593,7 @@ void sample_grammar(FileFormat file_format, int32_t n_vocab, llama_token_data_ar
for (const auto & reject : rejects) { for (const auto & reject : rejects) {
candidates->data[reject.index].logit = -INFINITY; candidates->data[reject.index].logit = -INFINITY;
} }
auto first = candidates->data; auto first = candidates->data;
auto last = first + candidates->size; auto last = first + candidates->size;
last = std::remove_if(first, last, last = std::remove_if(first, last,
@ -1650,25 +1650,19 @@ const std::vector<samplers> & sampler_order, llama_grammar * grammar, float dyna
//dry always first as logits cannot be resorted //dry always first as logits cannot be resorted
sample_dry(n_ctx, dry_penalty_last_n, dry_multiplier, dry_base, dry_allowed_length, dry_sequence_breakers, &candidates_p); sample_dry(n_ctx, dry_penalty_last_n, dry_multiplier, dry_base, dry_allowed_length, dry_sequence_breakers, &candidates_p);
//prefilter to top 3k tokens for improved speed //prefilter to top 3k tokens for improved speed
bool use_grammar = grammar != nullptr; bool use_grammar = grammar != nullptr;
size_t n_pre_cull = candidates_p.size; size_t n_pre_cull = candidates_p.size;
sample_top_k(&candidates_p, 3000);
if (use_grammar) {
(debugmode == 1 && printf("\nGrammar sampling %zu candidates.\n", candidates_p.size));
sample_grammar(file_format, n_vocab, &candidates_p, grammar);
(debugmode == 1 && printf("\nGrammar returned %zu candidates.\n", candidates_p.size));
sample_top_k(&candidates_p, 3000);
if (use_grammar) {
sample_grammar(file_format, n_vocab, &candidates_p, grammar);
// if top_k 3000 doesn't contain a valid candidate for this grammar, try again pre-cull // if top_k 3000 doesn't contain a valid candidate for this grammar, try again pre-cull
if (candidates_p.size <= 0) { if (candidates_p.size <= 0) {
candidates_p.size = n_pre_cull; candidates_p.size = n_pre_cull;
(debugmode == 1 && printf("\nRe-sampling grammar with %zu pre-cull tokens.\n", candidates_p.size));
sample_grammar(file_format, n_vocab, &candidates_p, grammar); sample_grammar(file_format, n_vocab, &candidates_p, grammar);
(debugmode == 1 && printf("\nGrammar returned %zu candidates.\n", candidates_p.size));
sample_top_k(&candidates_p, 3000); sample_top_k(&candidates_p, 3000);
} }
} }
@ -3960,7 +3954,6 @@ generation_outputs gpttype_generate(const generation_inputs inputs)
} }
if (grammar != nullptr) { if (grammar != nullptr) {
(debugmode == 1 && printf("\nGrammar attempting to accept token...\n"));
grammar_accept_token(file_format, n_vocab, grammar, id); grammar_accept_token(file_format, n_vocab, grammar, id);
} }