Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

rerank : use [SEP] token instead of [BOS] #9737

Merged
merged 4 commits into from
Oct 5, 2024
Merged

rerank : use [SEP] token instead of [BOS] #9737

Show file tree
Hide file tree
Changes from 3 commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
1ba3df3
rerank : use [SEP] token instead of [BOS]
ggerganov Oct 4, 2024
9e897d4
common : sanity check for non-NULL tokens
ggerganov Oct 4, 2024
e51973f
ci : adjust rank score interval
ggerganov Oct 4, 2024
6ada2e7
ci : add shebang to run.sh
ggerganov Oct 5, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 3 additions & 2 deletions ci/run.sh
View file
Open in desktop
Copy link
Contributor

@arch-btw arch-btw Oct 5, 2024 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Hi @ggerganov , since you're working on this file, I just wanted to point out that the very first line:

#/bin/bash

Might be missing the ! for the shebang:

#!/bin/bash

Original file line number Diff line number Diff line change
Expand Up @@ -751,7 +751,8 @@ function gg_run_rerank_tiny {

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
# rerank score 0: 0.029
Expand All @@ -774,7 +775,7 @@ function gg_run_rerank_tiny {

check_score "rerank score 0" "$(cat $OUT/${ci}-rk-f16.log | grep "rerank score 0")" "0.00" "0.05" | tee -a $OUT/${ci}-rk-f16.log
check_score "rerank score 1" "$(cat $OUT/${ci}-rk-f16.log | grep "rerank score 1")" "0.00" "0.05" | tee -a $OUT/${ci}-rk-f16.log
check_score "rerank score 2" "$(cat $OUT/${ci}-rk-f16.log | grep "rerank score 2")" "0.10" "0.15" | tee -a $OUT/${ci}-rk-f16.log
check_score "rerank score 2" "$(cat $OUT/${ci}-rk-f16.log | grep "rerank score 2")" "0.10" "0.30" | tee -a $OUT/${ci}-rk-f16.log

set +e
}
Expand Down
30 changes: 29 additions & 1 deletion common/common.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -838,6 +838,31 @@ struct llama_init_result llama_init_from_gpt_params(gpt_params & params) {
return iparams;
}

if (params.reranking) {
bool ok = true;

if (llama_token_bos(model) == LLAMA_TOKEN_NULL) {
LOG_WRN("%s: warning: model does not have a BOS token, reranking will not work\n", __func__);
ok = false;
}

if (llama_token_eos(model) == LLAMA_TOKEN_NULL) {
LOG_WRN("%s: warning: model does not have an EOS token, reranking will not work\n", __func__);
ok = false;
}

if (llama_token_sep(model) == LLAMA_TOKEN_NULL) {
LOG_WRN("%s: warning: model does not have a SEP token, reranking will not work\n", __func__);
ok = false;
}

if (!ok) {
llama_free_model(model);

return iparams;
}
}

auto cparams = llama_context_params_from_gpt_params(params);

llama_context * lctx = llama_new_context_with_model(model, cparams);
Expand All @@ -855,6 +880,7 @@ struct llama_init_result llama_init_from_gpt_params(gpt_params & params) {
if (cvec.n_embd == -1) {
llama_free(lctx);
llama_free_model(model);

return iparams;
}

Expand All @@ -867,6 +893,7 @@ struct llama_init_result llama_init_from_gpt_params(gpt_params & params) {
if (err) {
llama_free(lctx);
llama_free_model(model);

return iparams;
}
}
Expand All @@ -889,7 +916,7 @@ struct llama_init_result llama_init_from_gpt_params(gpt_params & params) {
llama_lora_adapters_apply(lctx, iparams.lora_adapters);
}

if (params.sparams.ignore_eos && llama_token_eos(model) == -1) {
if (params.sparams.ignore_eos && llama_token_eos(model) == LLAMA_TOKEN_NULL) {
LOG_WRN("%s: warning: model does not have an EOS token, ignoring --ignore-eos\n", __func__);
params.sparams.ignore_eos = false;
}
Expand Down Expand Up @@ -930,6 +957,7 @@ struct llama_init_result llama_init_from_gpt_params(gpt_params & params) {

iparams.model = model;
iparams.context = lctx;

return iparams;
}

Expand Down
4 changes: 2 additions & 2 deletions examples/server/server.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2027,15 +2027,15 @@ struct server_context {
continue;
}

// prompt: <s>query</s><s>doc</s>
// prompt: [BOS]query[EOS][SEP]doc[EOS]
prompt_tokens.clear();
prompt_tokens.push_back(llama_token_bos(model));
{
const auto part = tokenize(slot.prompt[0], false);
prompt_tokens.insert(prompt_tokens.end(), part.begin(), part.end());
}
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);
prompt_tokens.insert(prompt_tokens.end(), part.begin(), part.end());
Expand Down
18 changes: 9 additions & 9 deletions src/llama-vocab.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -40,17 +40,17 @@ struct llama_vocab {
id special_bos_id = 1;
id special_eos_id = 2;
id special_unk_id = 0;
id special_sep_id = -1;
id special_pad_id = -1;
id special_cls_id = -1;
id special_mask_id = -1;
id special_sep_id = LLAMA_TOKEN_NULL;
id special_pad_id = LLAMA_TOKEN_NULL;
id special_cls_id = LLAMA_TOKEN_NULL;
id special_mask_id = LLAMA_TOKEN_NULL;

id linefeed_id = 13;
id special_prefix_id = -1;
id special_suffix_id = -1;
id special_middle_id = -1;
id special_eot_id = -1; // TODO: move above after "eos_id", and here add "file separator" token
id special_eom_id = -1;
id special_prefix_id = LLAMA_TOKEN_NULL;
id special_suffix_id = LLAMA_TOKEN_NULL;
id special_middle_id = LLAMA_TOKEN_NULL;
id special_eot_id = LLAMA_TOKEN_NULL; // TODO: move above after "eos_id", and here add "file separator" token
id special_eom_id = LLAMA_TOKEN_NULL;

// set of all tokens that cause "end of generation"
std::set<id> special_eog_ids;
Expand Down
2 changes: 1 addition & 1 deletion src/llama.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2412,7 +2412,7 @@ struct llama_hparams {

// needed by encoder-decoder models (e.g. T5, FLAN-T5)
// ref: https://github.com/ggerganov/llama.cpp/pull/8141
llama_token dec_start_token_id = -1;
llama_token dec_start_token_id = LLAMA_TOKEN_NULL;

enum llama_pooling_type pooling_type = LLAMA_POOLING_TYPE_NONE;
enum llama_rope_type rope_type = LLAMA_ROPE_TYPE_NONE;
Expand Down
Loading