8000 server: --offline mode by ochafik · Pull Request #13804 · ggml-org/llama.cpp · GitHub
[go: up one dir, main page]

Skip to content

server: --offline mode #13804

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.

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

Merged
merged 11 commits into from
May 26, 2025
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
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
206 changes: 113 additions & 93 deletions common/arg.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -242,33 +242,7 @@ 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) {
// Initialize libcurl
curl_ptr curl(curl_easy_init(), &curl_easy_cleanup);
curl_slist_ptr http_headers;
if (!curl) {
LOG_ERR("%s: error initializing libcurl\n", __func__);
return false;
}

// Set the URL, allow to follow http redirection
curl_easy_setopt(curl.get(), CURLOPT_URL, url.c_str());
curl_easy_setopt(curl.get(), CURLOPT_FOLLOWLOCATION, 1L);

http_headers.ptr = curl_slist_append(http_headers.ptr, "User-Agent: llama-cpp");
// Check if hf-token or bearer-token was specified
if (!bearer_token.empty()) {
std::string auth_header = "Authorization: Bearer " + bearer_token;
http_headers.ptr = curl_slist_append(http_headers.ptr, auth_header.c_str());
}
curl_easy_setopt(curl.get(), CURLOPT_HTTPHEADER, http_headers.ptr);

#if defined(_WIN32)
// CURLSSLOPT_NATIVE_CA tells libcurl to use standard certificate store of
// operating system. Currently implemented under MS-Windows.
curl_easy_setopt(curl.get(), CURLOPT_SSL_OPTIONS, CURLSSLOPT_NATIVE_CA);
#endif

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);

Expand All @@ -279,6 +253,10 @@ static bool common_download_file_single(const std::string & url, const std::stri
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()) {
Expand All @@ -297,6 +275,10 @@ static bool common_download_file_single(const std::string & url, const std::stri
}
// 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;
Copy link
Collaborator
@ngxson ngxson May 26, 2025

Choose a reason for hiding this comment

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

This code branch will be trigger when file_exists == false && offline == true, so technically one of the condition inside if (offline) below now become dead code.

Instead, you can move the if (offline) to above of if (file_exists). So if file exist, we completely skip reading metadata

Copy link
Collaborator Author

Choose a reason for hiding this comment

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

ah ofc, rearranged to skip, thanks!

}
LOG_INF("%s: no previous model file found %s\n", __func__, path.c_str());
}

Expand All @@ -310,50 +292,73 @@ static bool common_download_file_single(const std::string & url, const std::stri
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;
// Initialize libcurl
curl_ptr curl(curl_easy_init(), &curl_easy_cleanup);
curl_slist_ptr http_headers;
if (!curl) {
LOG_ERR("%s: error initializing libcurl\n", __func__);
return false;
}

// Set the URL, allow to follow http redirection
curl_easy_setopt(curl.get(), CURLOPT_URL, url.c_str());
curl_easy_setopt(curl.get(), CURLOPT_FOLLOWLOCATION, 1L);

static std::regex header_regex("([^:]+): (.*)\r\n");
static std::regex etag_regex("ETag", std::regex_constants::icase);
static std::regex last_modified_regex("Last-Modified", std::regex_constants::icase);
http_headers.ptr = curl_slist_append(http_headers.ptr, "User-Agent: llama-cpp");
// Check if hf-token or bearer-token was specified
if (!bearer_token.empty()) {
std::string auth_header = "Authorization: Bearer " + bearer_token;
http_headers.ptr = curl_slist_append(http_headers.ptr, auth_header.c_str());
}
curl_easy_setopt(curl.get(), CURLOPT_HTTPHEADER, http_headers.ptr);

std::string header(buffer, n_items);
std::smatch match;
if (std::regex_match(header, match, header_regex)) {
const std::string & key = match[1];
const std::string & value = match[2];
if (std::regex_match(key, match, etag_regex)) {
headers->etag = value;
} else if (std::regex_match(key, match, last_modified_regex)) {
headers->last_modified = value;
}
}
return n_items;
};
#if defined(_WIN32)
// CURLSSLOPT_NATIVE_CA tells libcurl to use standard certificate store of
// operating system. Currently implemented under MS-Windows.
curl_easy_setopt(curl.get(), CURLOPT_SSL_OPTIONS, CURLSSLOPT_NATIVE_CA);
#endif

curl_easy_setopt(curl.get(), CURLOPT_NOBODY, 1L); // will trigger the HEAD verb
curl_easy_setopt(curl.get(), CURLOPT_NOPROGRESS, 1L); // hide head request progress
curl_easy_setopt(curl.get(), CURLOPT_HEADERFUNCTION, static_cast<CURLOPT_HEADERFUNCTION_PTR>(header_callback));
curl_easy_setopt(curl.get(), CURLOPT_HEADERDATA, &headers);
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;

// we only allow retrying once for HEAD requests
// this is for the use case of using running offline (no internet), retrying can be annoying
bool was_perform_successful = curl_perform_with_retry(url, curl.get(), 1, 0, "HEAD");
if (!was_perform_successful) {
head_request_ok = false;
}
static std::regex header_regex("([^:]+): (.*)\r\n");
static std::regex etag_regex("ETag", std::regex_constants::icase);
static std::regex last_modified_regex("Last-Modified", std::regex_constants::icase);

long http_code = 0;
curl_easy_getinfo(curl.get(), CURLINFO_RESPONSE_CODE, &http_code);
if (http_code == 200) {
head_request_ok = true;
} else {
LOG_WRN("%s: HEAD invalid http status code received: %ld\n", __func__, http_code);
head_request_ok = false;
std::string header(buffer, n_items);
std::smatch match;
if (std::regex_match(header, match, header_regex)) {
const std::string & key = match[1];
const std::string & value = match[2];
if (std::regex_match(key, match, etag_regex)) {
headers->etag = value;
} else if (std::regex_match(key, match, last_modified_regex)) {
headers->last_modified = value;
}
}
return n_items;
};

curl_easy_setopt(curl.get(), CURLOPT_NOBODY, 1L); // will trigger the HEAD verb
curl_easy_setopt(curl.get(), CURLOPT_NOPROGRESS, 1L); // hide head request progress
curl_easy_setopt(curl.get(), CURLOPT_HEADERFUNCTION, static_cast<CURLOPT_HEADERFUNCTION_PTR>(header_callback));
curl_easy_setopt(curl.get(), CURLOPT_HEADERDATA, &headers);

// we only allow retrying once for HEAD requests
// this is for the use case of using running offline (no internet), retrying can be annoying
bool was_perform_successful = curl_perform_with_retry(url, curl.get(), 1, 0, "HEAD");
if (!was_perform_successful) {
head_request_ok = false;
}

long http_code = 0;
curl_easy_getinfo(curl.get(), CURLINFO_RESPONSE_CODE, &http_code);
if (http_code == 200) {
head_request_ok = true;
} else {
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
Expand Down Expand Up @@ -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));
}

Expand All @@ -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;
}

Expand Down Expand Up @@ -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;
Expand Down Expand Up @@ -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];
Expand Down Expand Up @@ -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;
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
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 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;
Expand Down Expand Up @@ -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 {};
}
Expand All @@ -742,15 +754,16 @@ 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
{
if (!model.hf_repo.empty()) {
// 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
}
Expand Down Expand Up @@ -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);
Expand Down Expand Up @@ -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()) {
Expand All @@ -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) {
Expand Down Expand Up @@ -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.",
Expand Down
1 change: 1 addition & 0 deletions common/common.h
Original file line number Diff line number Diff line change
Expand Up @@ -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
Expand Down
Loading
0