Mirrors/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-25 13:58:46 +01:00
Code Issues Packages Projects Releases Wiki Activity

llama : UGM tokenizer init with UNK tokens instead of PAD

Browse Source
This commit is contained in:
Georgi Gerganov 2024-07-02 10:40:14 +03:00
parent 9eb5d5617d
commit 17bb0eaec3
No known key found for this signature in database
GPG Key ID: 449E073F9DC10735
1 changed files with 3 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
src/llama.cpp
View File

@ -14890,9 +14890,9 @@ struct llm_tokenizer_ugm {
size_t input_len = normalized.size();
// initialize score_sum to -FLT_MAX so it will be always lower than sums of token scores
std::vector<struct best_tokenization> tokenization_results(input_len + 1, {0, 0, -FLT_MAX});
std::vector<struct best_tokenization> tokenization_results(input_len + 1, {vocab.special_unk_id, 0, -FLT_MAX});
// at the beginning tokenization score is zero
tokenization_results[0] = { 0, 0, 0 };
tokenization_results[0] = { vocab.special_unk_id, 0, 0 };
for (size_t input_offset = 0; input_offset < input_len;) {
size_t prefix_offset = input_offset;
@ -14912,7 +14912,7 @@ struct llm_tokenizer_ugm {
single_codepoint_token_found = true;
}
llama_token token_id = node->value;
const auto &token_data = vocab.id_to_token[token_id];
const auto & token_data = vocab.id_to_token[token_id];
// we set the user-defined token scores to 0 to make them more likely to be selected
// (normal token scores are log probabilities, so they are negative)