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

common : do not add null tokens during warmup

Browse source 
ggml-ci
This commit is contained in:
Georgi Gerganov 2024-09-07 21:32:43 +03:00
parent e32d0816ed
commit ba6a97c390
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
1 changed files with 7 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

9
common/common.cpp
View file

@ -2514,10 +2514,15 @@ struct llama_init_result llama_init_from_gpt_params(gpt_params & params) {
llama_token bos = llama_token_bos(model); llama_token bos = llama_token_bos(model);
llama_token eos = llama_token_eos(model); llama_token eos = llama_token_eos(model);
// some models (e.g. T5) don't have a BOS token // some models (e.g. T5) don't have a BOS token
if (bos != -1) { if (bos != LLAMA_TOKEN_NULL) {
tmp.push_back(bos); tmp.push_back(bos);
} }
tmp.push_back(eos); if (eos != LLAMA_TOKEN_NULL) {
tmp.push_back(eos);
}
if (tmp.empty()) {
tmp.push_back(0);
}
if (llama_model_has_encoder(model)) { if (llama_model_has_encoder(model)) {
llama_encode(lctx, llama_batch_get_one(tmp.data(), tmp.size(), 0, 0)); llama_encode(lctx, llama_batch_get_one(tmp.data(), tmp.size(), 0, 0));