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

update common.cpp

Browse source
This commit is contained in:
slaren 2024-10-09 13:46:01 +02:00
parent cd097710d7
commit 672438dce1
3 changed files with 17 additions and 17 deletions
Download patch file Download diff file Expand all files Collapse all files

2
common/arg.cpp
View file

@ -1876,7 +1876,7 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
"if suffix/prefix are specified, template will be disabled\n" "if suffix/prefix are specified, template will be disabled\n"
"only commonly used templates are accepted:\nhttps://github.com/ggerganov/llama.cpp/wiki/Templates-supported-by-llama_chat_apply_template", "only commonly used templates are accepted:\nhttps://github.com/ggerganov/llama.cpp/wiki/Templates-supported-by-llama_chat_apply_template",
[](gpt_params & params, const std::string & value) { [](gpt_params & params, const std::string & value) {
if (!llama_chat_verify_template(value)) { if (!common_chat_verify_template(value)) {
throw std::runtime_error(format( throw std::runtime_error(format(
"error: the supplied chat template is not supported: %s\n" "error: the supplied chat template is not supported: %s\n"
"note: llama.cpp does not use jinja parser, we only support commonly used templates\n", "note: llama.cpp does not use jinja parser, we only support commonly used templates\n",

28
common/common.cpp
View file

@ -876,7 +876,7 @@ struct common_init_result llama_init_from_gpt_params(gpt_params & params) {
if (params.control_vector_layer_start <= 0) params.control_vector_layer_start = 1; if (params.control_vector_layer_start <= 0) params.control_vector_layer_start = 1;
if (params.control_vector_layer_end <= 0) params.control_vector_layer_end = llama_n_layer(model); if (params.control_vector_layer_end <= 0) params.control_vector_layer_end = llama_n_layer(model);
const auto cvec = llama_control_vector_load(params.control_vectors); const auto cvec = common_control_vector_load(params.control_vectors);
if (cvec.n_embd == -1) { if (cvec.n_embd == -1) {
llama_free(lctx); llama_free(lctx);
llama_free_model(model); llama_free_model(model);
@ -1112,7 +1112,7 @@ static bool curl_perform_with_retry(const std::string& url, CURL* curl, int max_
return false; return false;
} }
static bool llama_download_file(const std::string & url, const std::string & path, const std::string & hf_token) { static bool common_download_file(const std::string & url, const std::string & path, const std::string & hf_token) {
// Initialize libcurl // Initialize libcurl
std::unique_ptr<CURL, decltype(&curl_easy_cleanup)> curl(curl_easy_init(), &curl_easy_cleanup); std::unique_ptr<CURL, decltype(&curl_easy_cleanup)> curl(curl_easy_init(), &curl_easy_cleanup);
@ -1182,15 +1182,15 @@ static bool llama_download_file(const std::string & url, const std::string & pat
} }
// Send a HEAD request to retrieve the etag and last-modified headers // Send a HEAD request to retrieve the etag and last-modified headers
struct llama_load_model_from_url_headers { struct common_load_model_from_url_headers {
std::string etag; std::string etag;
std::string last_modified; std::string last_modified;
}; };
llama_load_model_from_url_headers headers; common_load_model_from_url_headers headers;
{ {
typedef size_t(*CURLOPT_HEADERFUNCTION_PTR)(char *, size_t, size_t, void *); typedef size_t(*CURLOPT_HEADERFUNCTION_PTR)(char *, size_t, size_t, void *);
auto header_callback = [](char * buffer, size_t /*size*/, size_t n_items, void * userdata) -> size_t { auto header_callback = [](char * buffer, size_t /*size*/, size_t n_items, void * userdata) -> size_t {
llama_load_model_from_url_headers *headers = (llama_load_model_from_url_headers *) userdata; common_load_model_from_url_headers *headers = (common_load_model_from_url_headers *) userdata;
static std::regex header_regex("([^:]+): (.*)\r\n"); static std::regex header_regex("([^:]+): (.*)\r\n");
static std::regex etag_regex("ETag", std::regex_constants::icase); static std::regex etag_regex("ETag", std::regex_constants::icase);
@ -1326,7 +1326,7 @@ static bool llama_download_file(const std::string & url, const std::string & pat
return true; return true;
} }
struct llama_model * llama_load_model_from_url( struct llama_model * common_load_model_from_url(
const char * model_url, const char * model_url,
const char * path_model, const char * path_model,
const char * hf_token, const char * hf_token,
@ -1337,7 +1337,7 @@ struct llama_model * llama_load_model_from_url(
return NULL; return NULL;
} }
if (!llama_download_file(model_url, path_model, hf_token)) { if (!common_download_file(model_url, path_model, hf_token)) {
return NULL; return NULL;
} }
@ -1390,7 +1390,7 @@ struct llama_model * llama_load_model_from_url(
char split_url[LLAMA_CURL_MAX_URL_LENGTH] = {0}; char split_url[LLAMA_CURL_MAX_URL_LENGTH] = {0};
llama_split_path(split_url, sizeof(split_url), split_url_prefix, download_idx, n_split); llama_split_path(split_url, sizeof(split_url), split_url_prefix, download_idx, n_split);
return llama_download_file(split_url, split_path, hf_token); return common_download_file(split_url, split_path, hf_token);
}, idx)); }, idx));
} }
@ -1405,7 +1405,7 @@ struct llama_model * llama_load_model_from_url(
return llama_load_model_from_file(path_model, params); return llama_load_model_from_file(path_model, params);
} }
struct llama_model * llama_load_model_from_hf( struct llama_model * common_load_model_from_hf(
const char * repo, const char * repo,
const char * model, const char * model,
const char * path_model, const char * path_model,
@ -1425,7 +1425,7 @@ struct llama_model * llama_load_model_from_hf(
model_url += "/resolve/main/"; model_url += "/resolve/main/";
model_url += model; model_url += model;
return llama_load_model_from_url(model_url.c_str(), path_model, hf_token, params); return common_load_model_from_url(model_url.c_str(), path_model, hf_token, params);
} }
#else #else
@ -1545,7 +1545,7 @@ std::string common_detokenize(llama_context * ctx, const std::vector<llama_token
// Chat template utils // Chat template utils
// //
bool llama_chat_verify_template(const std::string & tmpl) { bool common_chat_verify_template(const std::string & tmpl) {
llama_chat_message chat[] = {{"user", "test"}}; llama_chat_message chat[] = {{"user", "test"}};
int res = llama_chat_apply_template(nullptr, tmpl.c_str(), chat, 1, true, nullptr, 0); int res = llama_chat_apply_template(nullptr, tmpl.c_str(), chat, 1, true, nullptr, 0);
return res >= 0; return res >= 0;
@ -1765,7 +1765,7 @@ float common_embd_similarity_cos(const float * embd1, const float * embd2, int n
// Control vector utils // Control vector utils
// //
static common_control_vector_data llama_control_vector_load_one(const common_control_vector_load_info & load_info) { static common_control_vector_data common_control_vector_load_one(const common_control_vector_load_info & load_info) {
common_control_vector_data result = { -1, {} }; common_control_vector_data result = { -1, {} };
ggml_context * ctx = nullptr; ggml_context * ctx = nullptr;
@ -1850,11 +1850,11 @@ static common_control_vector_data llama_control_vector_load_one(const common_con
return result; return result;
} }
common_control_vector_data llama_control_vector_load(const std::vector<common_control_vector_load_info> & load_infos) { common_control_vector_data common_control_vector_load(const std::vector<common_control_vector_load_info> & load_infos) {
common_control_vector_data result = { -1, {} }; common_control_vector_data result = { -1, {} };
for (const auto & info : load_infos) { for (const auto & info : load_infos) {
auto cur = llama_control_vector_load_one(info); auto cur = common_control_vector_load_one(info);
if (cur.n_embd == -1) { if (cur.n_embd == -1) {
result.n_embd = -1; result.n_embd = -1;

4
common/common.h
View file

@ -477,7 +477,7 @@ struct common_chat_msg {
}; };
// Check if the template supplied via "--chat-template" is supported or not. Returns true if it's valid // Check if the template supplied via "--chat-template" is supported or not. Returns true if it's valid
bool llama_chat_verify_template(const std::string & tmpl); bool common_chat_verify_template(const std::string & tmpl);
// CPP wrapper for llama_chat_apply_template // CPP wrapper for llama_chat_apply_template
// If the built-in template is not supported, we default to chatml // If the built-in template is not supported, we default to chatml
@ -535,7 +535,7 @@ struct common_control_vector_load_info {
// Load control vectors, scale each by strength, and add them together. // Load control vectors, scale each by strength, and add them together.
// On error, returns {-1, empty} // On error, returns {-1, empty}
common_control_vector_data llama_control_vector_load(const std::vector<common_control_vector_load_info> & load_infos); common_control_vector_data common_control_vector_load(const std::vector<common_control_vector_load_info> & load_infos);
// //
// Split utils // Split utils