tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-27 12:05:03 +00:00
Code Issues Packages Projects Releases Wiki Activity

server: --offline mode (#13804)

Browse Source 
* server: --offline mode (env: LLAMA_OFFLINE)

---------

Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com>
This commit is contained in:
Olivier Chafik
2025-05-26 14:34:27 -07:00
committed by GitHub
parent a26c4cc11e
commit cdf94a1802
2 changed files with 127 additions and 106 deletions
Download Patch File Download Diff File Expand all files Collapse all files

158
common/arg.cpp
View File

@ -242,7 +242,56 @@ static bool curl_perform_with_retry(const std::string & url, CURL * curl, int ma
}
// download one single file from remote URL to local path
static bool common_download_file_single(const std::string & url, const std::string & path, const std::string & bearer_token) {
static bool common_download_file_single(const std::string & url, const std::string & path, const std::string & bearer_token, bool offline) {
// Check if the file already exists locally
auto file_exists = std::filesystem::exists(path);
// If the file exists, check its JSON metadata companion file.
std::string metadata_path = path + ".json";
nlohmann::json metadata; // TODO @ngxson : get rid of this json, use regex instead
std::string etag;
std::string last_modified;
if (file_exists) {
if (offline) {
LOG_INF("%s: using cached file (offline mode): %s\n", __func__, path.c_str());
return true; // skip verification/downloading
}
// Try and read the JSON metadata file (note: stream autoclosed upon exiting this block).
std::ifstream metadata_in(metadata_path);
if (metadata_in.good()) {
try {
metadata_in >> metadata;
LOG_DBG("%s: previous metadata file found %s: %s\n", __func__, metadata_path.c_str(), metadata.dump().c_str());
if (metadata.contains("etag") && metadata.at("etag").is_string()) {
etag = metadata.at("etag");
}
if (metadata.contains("lastModified") && metadata.at("lastModified").is_string()) {
last_modified = metadata.at("lastModified");
}
} catch (const nlohmann::json::exception & e) {
LOG_ERR("%s: error reading metadata file %s: %s\n", __func__, metadata_path.c_str(), e.what());
}
}
// if we cannot open the metadata file, we assume that the downloaded file is not valid (etag and last-modified are left empty, so we will download it again)
} else {
if (offline) {
LOG_ERR("%s: required file is not available in cache (offline mode): %s\n", __func__, path.c_str());
return false;
}
LOG_INF("%s: no previous model file found %s\n", __func__, path.c_str());
}
// Send a HEAD request to retrieve the etag and last-modified headers
struct common_load_model_from_url_headers {
std::string etag;
std::string last_modified;
};
common_load_model_from_url_headers headers;
bool head_request_ok = false;
bool should_download = !file_exists; // by default, we should download if the file does not exist
// Initialize libcurl
curl_ptr curl(curl_easy_init(), &curl_easy_cleanup);
curl_slist_ptr http_headers;
@ -269,49 +318,6 @@ static bool common_download_file_single(const std::string & url, const std::stri
curl_easy_setopt(curl.get(), CURLOPT_SSL_OPTIONS, CURLSSLOPT_NATIVE_CA);
#endif
// Check if the file already exists locally
auto file_exists = std::filesystem::exists(path);
// If the file exists, check its JSON metadata companion file.
std::string metadata_path = path + ".json";
nlohmann::json metadata; // TODO @ngxson : get rid of this json, use regex instead
std::string etag;
std::string last_modified;
if (file_exists) {
// Try and read the JSON metadata file (note: stream autoclosed upon exiting this block).
std::ifstream metadata_in(metadata_path);
if (metadata_in.good()) {
try {
metadata_in >> metadata;
LOG_DBG("%s: previous metadata file found %s: %s\n", __func__, metadata_path.c_str(), metadata.dump().c_str());
if (metadata.contains("etag") && metadata.at("etag").is_string()) {
etag = metadata.at("etag");
}
if (metadata.contains("lastModified") && metadata.at("lastModified").is_string()) {
last_modified = metadata.at("lastModified");
}
} catch (const nlohmann::json::exception & e) {
LOG_ERR("%s: error reading metadata file %s: %s\n", __func__, metadata_path.c_str(), e.what());
}
}
// if we cannot open the metadata file, we assume that the downloaded file is not valid (etag and last-modified are left empty, so we will download it again)
} else {
LOG_INF("%s: no previous model file found %s\n", __func__, path.c_str());
}
// Send a HEAD request to retrieve the etag and last-modified headers
struct common_load_model_from_url_headers {
std::string etag;
std::string last_modified;
};
common_load_model_from_url_headers headers;
bool head_request_ok = false;
bool should_download = !file_exists; // by default, we should download if the file does not exist
// get ETag to see if the remote file has changed
{
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 {
common_load_model_from_url_headers * headers = (common_load_model_from_url_headers *) userdata;
@ -354,7 +360,6 @@ static bool common_download_file_single(const std::string & url, const std::stri
LOG_WRN("%s: HEAD invalid http status code received: %ld\n", __func__, http_code);
head_request_ok = false;
}
}
// if head_request_ok is false, we don't have the etag or last-modified headers
// we leave should_download as-is, which is true if the file does not exist
@ -460,12 +465,12 @@ static bool common_download_file_single(const std::string & url, const std::stri
// download multiple files from remote URLs to local paths
// the input is a vector of pairs <url, path>
static bool common_download_file_multiple(const std::vector<std::pair<std::string, std::string>> & urls, const std::string & bearer_token) {
static bool common_download_file_multiple(const std::vector<std::pair<std::string, std::string>> & urls, const std::string & bearer_token, bool offline) {
// Prepare download in parallel
std::vector<std::future<bool>> futures_download;
for (auto const & item : urls) {
futures_download.push_back(std::async(std::launch::async, [bearer_token](const std::pair<std::string, std::string> & it) -> bool {
return common_download_file_single(it.first, it.second, bearer_token);
futures_download.push_back(std::async(std::launch::async, [bearer_token, offline](const std::pair<std::string, std::string> & it) -> bool {
return common_download_file_single(it.first, it.second, bearer_token, offline);
}, item));
}
@ -481,14 +486,15 @@ static bool common_download_file_multiple(const std::vector<std::pair<std::strin
static bool common_download_model(
const common_params_model & model,
const std::string & bearer_token) {
const std::string & bearer_token,
bool offline) {
// Basic validation of the model.url
if (model.url.empty()) {
LOG_ERR("%s: invalid model url\n", __func__);
return false;
}
if (!common_download_file_single(model.url, model.path, bearer_token)) {
if (!common_download_file_single(model.url, model.path, bearer_token, offline)) {
return false;
}
@ -547,7 +553,7 @@ static bool common_download_model(
}
// Download in parallel
common_download_file_multiple(urls, bearer_token);
common_download_file_multiple(urls, bearer_token, offline);
}
return true;
@ -608,7 +614,7 @@ std::pair<long, std::vector<char>> common_remote_get_content(const std::string &
*
* Note: we use the Ollama-compatible HF API, but not using the blobId. Instead, we use the special "ggufFile" field which returns the value for "hf_file". This is done to be backward-compatible with existing cache files.
*/
static struct common_hf_file_res common_get_hf_file(const std::string & hf_repo_with_tag, const std::string & bearer_token) {
static struct common_hf_file_res common_get_hf_file(const std::string & hf_repo_with_tag, const std::string & bearer_token, bool offline) {
auto parts = string_split<std::string>(hf_repo_with_tag, ':');
std::string tag = parts.size() > 1 ? parts.back() : "latest";
std::string hf_repo = parts[0];
@ -638,20 +644,25 @@ static struct common_hf_file_res common_get_hf_file(const std::string & hf_repo_
long res_code = 0;
std::string res_str;
bool use_cache = false;
if (!offline) {
try {
auto res = common_remote_get_content(url, params);
res_code = res.first;
res_str = std::string(res.second.data(), res.second.size());
} catch (const std::exception & e) {
LOG_WRN("error: failed to get manifest: %s\n", e.what());
LOG_WRN("try reading from cache\n");
// try to read from cache
try {
LOG_WRN("error: failed to get manifest at %s: %s\n", url.c_str(), e.what());
}
}
if (res_code == 0) {
if (std::filesystem::exists(cached_response_path)) {
LOG_WRN("trying to read manifest from cache: %s\n", cached_response_path.c_str());
res_str = read_file(cached_response_path);
res_code = 200;
use_cache = true;
} catch (const std::exception & e) {
throw std::runtime_error("error: failed to get manifest (check your internet connection)");
} else {
throw std::runtime_error(
offline ? "error: failed to get manifest (offline mode)"
: "error: failed to get manifest (check your internet connection)");
}
}
std::string ggufFile;
@ -698,24 +709,25 @@ bool common_has_curl() {
return false;
}
static bool common_download_file_single(const std::string &, const std::string &, const std::string &) {
static bool common_download_file_single(const std::string &, const std::string &, const std::string &, bool) {
LOG_ERR("error: built without CURL, cannot download model from internet\n");
return false;
}
static bool common_download_file_multiple(const std::vector<std::pair<std::string, std::string>> &, const std::string &) {
static bool common_download_file_multiple(const std::vector<std::pair<std::string, std::string>> &, const std::string &, bool) {
LOG_ERR("error: built without CURL, cannot download model from the internet\n");
return false;
}
static bool common_download_model(
const common_params_model &,
const std::string &) {
const std::string &,
bool) {
LOG_ERR("error: built without CURL, cannot download model from the internet\n");
return false;
}
static struct common_hf_file_res common_get_hf_file(const std::string &, const std::string &) {
static struct common_hf_file_res common_get_hf_file(const std::string &, const std::string &, bool) {
LOG_ERR("error: built without CURL, cannot download model from the internet\n");
return {};
}
@ -742,7 +754,8 @@ struct handle_model_result {
static handle_model_result common_params_handle_model(
struct common_params_model & model,
const std::string & bearer_token,
const std::string & model_path_default) {
const std::string & model_path_default,
bool offline) {
handle_model_result result;
// handle pre-fill default model path and url based on hf_repo and hf_file
{
@ -750,7 +763,7 @@ static handle_model_result common_params_handle_model(
// short-hand to avoid specifying --hf-file -> default it to --model
if (model.hf_file.empty()) {
if (model.path.empty()) {
auto auto_detected = common_get_hf_file(model.hf_repo, bearer_token);
auto auto_detected = common_get_hf_file(model.hf_repo, bearer_token, offline);
if (auto_detected.repo.empty() || auto_detected.ggufFile.empty()) {
exit(1); // built without CURL, error message already printed
}
@ -791,7 +804,7 @@ static handle_model_result common_params_handle_model(
// then, download it if needed
if (!model.url.empty()) {
bool ok = common_download_model(model, bearer_token);
bool ok = common_download_model(model, bearer_token, offline);
if (!ok) {
LOG_ERR("error: failed to download model from %s\n", model.url.c_str());
exit(1);
@ -934,7 +947,7 @@ static bool common_params_parse_ex(int argc, char ** argv, common_params_context
// handle model and download
{
auto res = common_params_handle_model(params.model, params.hf_token, DEFAULT_MODEL_PATH);
auto res = common_params_handle_model(params.model, params.hf_token, DEFAULT_MODEL_PATH, params.offline);
if (params.no_mmproj) {
params.mmproj = {};
} else if (res.found_mmproj && params.mmproj.path.empty() && params.mmproj.url.empty()) {
@ -944,12 +957,12 @@ static bool common_params_parse_ex(int argc, char ** argv, common_params_context
// only download mmproj if the current example is using it
for (auto & ex : mmproj_examples) {
if (ctx_arg.ex == ex) {
common_params_handle_model(params.mmproj, params.hf_token, "");
common_params_handle_model(params.mmproj, params.hf_token, "", params.offline);
break;
}
}
common_params_handle_model(params.speculative.model, params.hf_token, "");
common_params_handle_model(params.vocoder.model, params.hf_token, "");
common_params_handle_model(params.speculative.model, params.hf_token, "", params.offline);
common_params_handle_model(params.vocoder.model, params.hf_token, "", params.offline);
}
if (params.escape) {
@ -2996,6 +3009,13 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
common_log_set_verbosity_thold(INT_MAX);
}
));
add_opt(common_arg(
{"--offline"},
"Offline mode: forces use of cache, prevents network access",
[](common_params & params) {
params.offline = true;
}
).set_env("LLAMA_OFFLINE"));
add_opt(common_arg(
{"-lv", "--verbosity", "--log-verbosity"}, "N",
"Set the verbosity threshold. Messages with a higher verbosity will be ignored.",

1
common/common.h
View File

@ -291,6 +291,7 @@ struct common_params {
int32_t verbosity = 0;
int32_t control_vector_layer_start = -1; // layer range for control vector
int32_t control_vector_layer_end = -1; // layer range for control vector
bool offline = false;
int32_t ppl_stride = 0; // stride for perplexity calculations. If left at 0, the pre-existing approach will be used.
int32_t ppl_output_type = 0; // = 0 -> ppl output is as usual, = 1 -> ppl output is num_tokens, ppl, one per line