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

rerank : use [SEP] token instead of [BOS]

Browse source 
ggml-ci
This commit is contained in:
Georgi Gerganov 2024-10-04 11:54:32 +03:00
parent d5ed2b929d
commit 1ba3df3de5
No known key found for this signature in database
GPG key ID: BF970631944C16B7
2 changed files with 4 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

3
ci/run.sh
View file

@ -751,7 +751,8 @@ function gg_run_rerank_tiny {
model_f16="${path_models}/ggml-model-f16.gguf" model_f16="${path_models}/ggml-model-f16.gguf"
(time ./bin/llama-embedding --model ${model_f16} -p "what is panda?</s><s>hi\nwhat is panda?</s><s>it's a bear\nwhat is panda?</s><s>The giant panda (Ailuropoda melanoleuca), sometimes called a panda bear or simply panda, is a bear species endemic to China." --pooling rank --embd-normalize -1 --verbose-prompt) 2>&1 | tee -a $OUT/${ci}-rk-f16.log # for this model, the SEP token is "</s>"
(time ./bin/llama-embedding --model ${model_f16} -p "what is panda?</s></s>hi\nwhat is panda?</s></s>it's a bear\nwhat is panda?</s></s>The giant panda (Ailuropoda melanoleuca), sometimes called a panda bear or simply panda, is a bear species endemic to China." --pooling rank --embd-normalize -1 --verbose-prompt) 2>&1 | tee -a $OUT/${ci}-rk-f16.log
# sample output # sample output
# rerank score 0: 0.029 # rerank score 0: 0.029

4
examples/server/server.cpp
View file

@ -2027,7 +2027,7 @@ struct server_context {
continue; continue;
} }
// prompt: <s>query</s><s>doc</s> // prompt: [BOS]query[EOS][SEP]doc[EOS]
prompt_tokens.clear(); prompt_tokens.clear();
prompt_tokens.push_back(llama_token_bos(model)); prompt_tokens.push_back(llama_token_bos(model));
{ {
@ -2035,7 +2035,7 @@ struct server_context {
prompt_tokens.insert(prompt_tokens.end(), part.begin(), part.end()); prompt_tokens.insert(prompt_tokens.end(), part.begin(), part.end());
} }
prompt_tokens.push_back(llama_token_eos(model)); prompt_tokens.push_back(llama_token_eos(model));
prompt_tokens.push_back(llama_token_bos(model)); prompt_tokens.push_back(llama_token_sep(model));
{ {
const auto part = tokenize(slot.prompt[1], false); const auto part = tokenize(slot.prompt[1], false);
prompt_tokens.insert(prompt_tokens.end(), part.begin(), part.end()); prompt_tokens.insert(prompt_tokens.end(), part.begin(), part.end());