tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-26 19:55:04 +00:00
Code Issues Packages Projects Releases Wiki Activity

llama : rework embeddings logic (#14208)

Browse Source 
* llama : rework embeddings logic

ggml-ci

* cont : fix rerank

ggml-ci

* cont : engrish [no ci]

* cont : fix rerank

ggml-ci

* server : support both embeddings and completions with single model

ggml-ci

* cont : avoid embeddings_org

ggml-ci
This commit is contained in:
Georgi Gerganov
2025-06-16 14:14:00 +03:00
committed by GitHub
parent 3ba0d843c6
commit d3e64b9f49
16 changed files with 159 additions and 114 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
common/arg.cpp
View File

@ -988,10 +988,6 @@ static bool common_params_parse_ex(int argc, char ** argv, common_params_context
params.tensor_buft_overrides.push_back({nullptr, nullptr});
}
if (params.reranking && params.embedding) {
throw std::invalid_argument("error: either --embedding or --reranking can be specified, but not both");
}
if (!params.chat_template.empty() && !common_chat_verify_template(params.chat_template, params.use_jinja)) {
throw std::runtime_error(string_format(
"error: the supplied chat template is not supported: %s%s\n",
@ -2747,9 +2743,10 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_EMBEDDINGS"));
add_opt(common_arg(
{"--reranking", "--rerank"},
string_format("enable reranking endpoint on server (default: %s)", params.reranking ? "enabled" : "disabled"),
string_format("enable reranking endpoint on server (default: %s)", "disabled"),
[](common_params & params) {
params.reranking = true;
params.embedding = true;
params.pooling_type = LLAMA_POOLING_TYPE_RANK;
}
).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_RERANKING"));
add_opt(common_arg(

62
common/common.cpp
View File

@ -897,34 +897,6 @@ struct common_init_result common_init_from_params(common_params & params) {
const llama_vocab * vocab = llama_model_get_vocab(model);
if (params.reranking) {
bool ok = true;
if (llama_vocab_bos(vocab) == LLAMA_TOKEN_NULL) {
LOG_WRN("%s: warning: vocab does not have a BOS token, reranking will not work\n", __func__);
ok = false;
}
bool has_eos = llama_vocab_eos(vocab) != LLAMA_TOKEN_NULL;
bool has_sep = llama_vocab_sep(vocab) != LLAMA_TOKEN_NULL;
if (!has_eos && !has_sep) {
LOG_WRN("%s: warning: vocab does not have an EOS token or SEP token, reranking will not work\n", __func__);
ok = false;
} else if (!has_eos) {
LOG_WRN("%s: warning: vocab does not have an EOS token, using SEP token as fallback\n", __func__);
} else if (!has_sep) {
LOG_WRN("%s: warning: vocab does not have a SEP token, reranking will not work\n", __func__);
ok = false;
}
if (!ok) {
llama_model_free(model);
return iparams;
}
}
auto cparams = common_context_params_to_llama(params);
llama_context * lctx = llama_init_from_model(model, cparams);
@ -966,6 +938,35 @@ struct common_init_result common_init_from_params(common_params & params) {
}
}
if (llama_pooling_type(lctx) == LLAMA_POOLING_TYPE_RANK) {
bool ok = true;
if (llama_vocab_bos(vocab) == LLAMA_TOKEN_NULL) {
LOG_WRN("%s: warning: vocab does not have a BOS token, reranking will not work\n", __func__);
ok = false;
}
bool has_eos = llama_vocab_eos(vocab) != LLAMA_TOKEN_NULL;
bool has_sep = llama_vocab_sep(vocab) != LLAMA_TOKEN_NULL;
if (!has_eos && !has_sep) {
LOG_WRN("%s: warning: vocab does not have an EOS token or SEP token, reranking will not work\n", __func__);
ok = false;
} else if (!has_eos) {
LOG_WRN("%s: warning: vocab does not have an EOS token, using SEP token as fallback\n", __func__);
} else if (!has_sep) {
LOG_WRN("%s: warning: vocab does not have a SEP token, reranking will not work\n", __func__);
ok = false;
}
if (!ok) {
llama_free(lctx);
llama_model_free(model);
return iparams;
}
}
// load and optionally apply lora adapters
for (auto & la : params.lora_adapters) {
llama_adapter_lora_ptr lora;
@ -1143,11 +1144,6 @@ struct llama_context_params common_context_params_to_llama(const common_params &
cparams.op_offload = !params.no_op_offload;
cparams.swa_full = params.swa_full;
if (params.reranking) {
cparams.embeddings = true;
cparams.pooling_type = LLAMA_POOLING_TYPE_RANK;
}
cparams.type_k = params.cache_type_k;
cparams.type_v = params.cache_type_v;

1
common/common.h
View File

@ -355,7 +355,6 @@ struct common_params {
int32_t embd_normalize = 2; // normalisation for embeddings (-1=none, 0=max absolute int16, 1=taxicab, 2=euclidean, >2=p-norm)
std::string embd_out = ""; // empty = default, "array" = [[],[]...], "json" = openai style, "json+" = same "json" + cosine similarity matrix
std::string embd_sep = "\n"; // separator of embeddings
bool reranking = false; // enable reranking support on server
// server params
int32_t port = 8080; // server listens on this network port