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

llama : Use token_to_id map find() method instead of iterating over all tokens.

Browse source
This commit is contained in:
Stanisław Szymczyk 2024-08-04 20:47:47 +02:00
parent 3878b397a9
commit 0b7211387e
1 changed files with 3 additions and 5 deletions
Download patch file Download diff file Expand all files Collapse all files

8
src/llama.cpp
View file

@ -5647,11 +5647,9 @@ static void llm_load_vocab(
// TODO: convert scripts should provide this token through the KV metadata LLAMA_KV_TOKENIZER_EOM_ID // TODO: convert scripts should provide this token through the KV metadata LLAMA_KV_TOKENIZER_EOM_ID
// for now, we apply this workaround to find the EOM token based on its text // for now, we apply this workaround to find the EOM token based on its text
if (vocab.special_eom_id == -1) { if (vocab.special_eom_id == -1) {
for (const auto & t : vocab.token_to_id) { const auto & t = vocab.token_to_id.find("<|eom_id|>");
if (t.first == "<|eom_id|>") { if (t != vocab.token_to_id.end()) {
vocab.special_eom_id = t.second; vocab.special_eom_id = t->second;
break;
}
} }
} }