LLM-Inference/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-08-12 22:31:11 +04:00
Code Issues Packages Projects Releases Wiki Activity
Files
f785fc9ea485e6cfdda129978310aa52939c3619
llama.cpp/tools/server/server.cpp
T
PascalGitHubXuan Son Nguyen
 4ae84dea27 server: add more tool isolation support (ssh remote + podman rootless) (#26774) 
* server: add an ssh transport to the tools runtime

--tools-runtime ssh:<target> runs the built-in tools on a remote host,
where target is whatever ssh already resolves, a user@host or a config
alias, so no credentials live in llama.cpp.

Only build_argv and upload differ from the docker transport: the remote
shell re-parses the command line, so the argv travels through
shell_quote_join, and files go over scp with the same quoting on the
remote path. Authentication is key-based and the host key must already
be trusted, since the tools run without a console and any prompt would
hang them.

The target is validated before use. The spec can reach us from the
x-tool-runtime header, and a leading dash would turn it into an ssh
option, which is enough to run a command back on the host.

Nothing is created and nothing is reclaimed, so an ssh spec goes
straight to the tool call instead of through the container runtime.

Note that this is remoting rather than isolation: the tools can do
whatever the target account can do, and the isolation is whatever runs
them on the far side.

* server: support podman in the tools runtime

docker and podman expose the same run, exec, cp and inspect verbs with the
same argument order, so a single implementation drives both and the engine
is carried by the spec prefix: podman:<image> and podman-container:<id> sit
next to the docker forms.

tools_io_docker becomes tools_io_container and the runtime spawner becomes
server_tools_container_runtime, both holding the client binary chosen at
parse time. A single parse_container_runtime() resolves every spec, so
adding another engine is one string in the table.

make_tools_io() now rejects the spawning forms. The spec also reaches it
from the x-tool-runtime header, which is client controlled, and only the
runtime that owns a container is allowed to create one: a tool call can
attach to a running container, nothing more.

* ./build/bin/llama-gen-docs

* server: simplify the tools runtime and drop the file copy step

A server_tools_runtime base with one virtual spec() replaces the
container runtime and the bare spec string that ssh needed next to it,
so server_tools is back to a single pointer and neither setup nor the
handler tests which of the two is set.

write_file used to spill its content into a temporary file on the host
and copy it in, because run_subprocess had no way to feed a child. It
now takes an optional stdin payload and creates the parent directory
and the file in a single round trip through a shell in the isolate.

That removes the upload virtual and both implementations: no more
container cp or scp, no second binary on the host, no sftp subsystem on
the target, no predictable temporary in a shared tmp, and none of the
content reaching an argv the remote shell re-parses. It also fixes
write_file over ssh, which never worked: scp speaks sftp and takes the
remote path literally, so quoting it kept the quotes in the file name.

Writing the payload before reading the output relies on the child
draining stdin as it goes, which holds for cat, its only user today.

* ./build/bin/llama-gen-docs

* server: harden the tools runtime against argv injection and a stdin stall

Validate the container id from x-tool-runtime and --tools-runtime the
same way the ssh target already is, so an id shaped like an option
(docker-container:--privileged) is rejected before it reaches the
engine's exec command line instead of running against a hardened
container. Feed the child's stdin after the watchdog is armed, so a
transport that stalls mid-write is terminated at the deadline rather
than blocking the request forever.

Cover both guards and fix the unknown-scheme test, which used ssh: as
its example and now names a real runtime.

* tests: exercise the tools runtime tests on podman as well as docker

Follow-up #26507. The container runtime drives docker and podman
through one implementation, so parametrize the availability helper,
the container fixture and the attach test on the engine, and cover
both engine prefixes in the container id injection test. Each engine
skips on its own when it is not installed.

The spawn cleanup test stays docker only: it recovers the spawned id
from the container hostname, which docker sets to the short id and
podman rootless does not guarantee. Podman keeps its coverage through
the attach path.

* server: release the container handle before respawning

Follow-up #26507. create() writes over the handle it is given, so a
respawn after the container died on its own leaked the pipes and the
process handle of the previous one.

* server: trim the tools runtime comments

* server: read tool output as raw bytes and harden the runtime on Windows

The stdout pipe is read with read() instead of fgets(), so a chunk
can hold any byte, including NUL, and still streams as soon as data
is available. Past the size cap the pipe keeps draining so the child
never blocks on a full pipe. Both pipe fds are forced to binary mode
on Windows, where the CRT defaults them to text mode and translates
line endings in both directions. Stdin is now always closed after
the feed: the child reads a deterministic EOF, and the Windows
docker and ssh clients stop outliving their command on a stdin pipe
that never closes.

The attach form of --tools-runtime has no lifecycle to own, so it
becomes a static target validated once at startup. This removes the
 subprocess that ran on every tool call and
serialized calls behind a mutex; a stopped container now surfaces
the engine's own error at exec time.

The cidfile path is passed as UTF-8, matching the encoding the
subprocess layer expects for the CreateProcessW command line, so
the spawn form works from a non-ASCII Windows profile.

The SIGPIPE note in server.cpp now names the tools runtime children
as well as the MCP ones.

* clean up comments

* less pollute global scope

* nits

* tests: name the container image after both engines

---------

Co-authored-by: Xuan Son Nguyen <son@huggingface.co>
2026-08-10 13:31:09 +02:00

538 lines
22 KiB
C++
Raw Blame History

#include "server-context.h"
#include "server-http.h"
#include "server-models.h"
#include "server-cors-proxy.h"
#include "server-stream.h"
#include "server-tools.h"
#include "arg.h"
#include "build-info.h"
#include "common.h"
#include "fit.h"
#include "llama.h"
#include "log.h"
#include <atomic>
#include <clocale>
#include <exception>
#include <signal.h>
#include <thread> // for std::thread::hardware_concurrency
#if defined(_WIN32)
#include <windows.h>
#endif
static std::function<void(int)> shutdown_handler;
static std::atomic_flag is_terminating = ATOMIC_FLAG_INIT;
static inline void signal_handler(int signal) {
if (is_terminating.test_and_set()) {
// in case it hangs, we can force terminate the server by hitting Ctrl+C twice
// this is for better developer experience, we can remove when the server is stable enough
fprintf(stderr, "Received second interrupt, terminating immediately.\n");
exit(1);
}
shutdown_handler(signal);
}
// satisfies -Wmissing-declarations (used by llama command)
int llama_server(int argc, char ** argv);
// to be used via CLI (argc / argv are used by router mode only)
int llama_server(common_params & params, int argc, char ** argv);
void llama_server_terminate();
void llama_server_terminate() {
if (shutdown_handler) {
shutdown_handler(0);
}
}
// wrapper function that handles exceptions and logs errors
// this is to make sure handler_t never throws exceptions; instead, it returns an error response
static server_http_context::handler_t ex_wrapper(server_http_context::handler_t func) {
return [func = std::move(func)](const server_http_req & req) -> server_http_res_ptr {
std::string message;
error_type error;
try {
return func(req);
} catch (const std::invalid_argument & e) {
// treat invalid_argument as invalid request (400)
error = ERROR_TYPE_INVALID_REQUEST;
message = e.what();
} catch (const std::exception & e) {
// treat other exceptions as server error (500)
error = ERROR_TYPE_SERVER;
message = e.what();
} catch (...) {
error = ERROR_TYPE_SERVER;
message = "unknown error";
}
auto res = std::make_unique<server_http_res>();
res->status = 500;
try {
json error_data = format_error_response(message, error);
res->status = json_value(error_data, "code", 500);
res->data = safe_json_to_str({{ "error", error_data }});
SRV_WRN("got exception: %s\n", res->data.c_str());
} catch (const std::exception & e) {
SRV_ERR("got another exception: %s | while handling exception: %s\n", e.what(), message.c_str());
res->data = "Internal Server Error";
}
return res;
};
}
int llama_server(int argc, char ** argv) {
std::setlocale(LC_NUMERIC, "C");
#ifndef _WIN32
// Ignore SIGPIPE so the server does not crash if a child (MCP server, tools runtime) exits while we are writing to its stdin
signal(SIGPIPE, SIG_IGN);
#endif
// own arguments required by this example
common_params params;
common_init();
// start the stream session manager GC right after common init, before any HTTP route can
// touch it. lifecycle is symmetric, stop_gc() runs in clean_up() before backend free
server_stream_session_manager_start();
if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_SERVER)) {
return 1;
}
llama_backend_init();
llama_numa_init(params.numa);
return llama_server(params, argc, argv);
}
int llama_server(common_params & params, int argc, char ** argv) {
bool is_run_by_cli = (argv == nullptr);
common_models_handler models_handler;
// note: router mode also accepts -hf remote-preset, so we need to check that first
if (!is_run_by_cli && !params.model.hf_repo.empty()) {
try {
models_handler = common_models_handler_init(params, LLAMA_EXAMPLE_SERVER);
if (common_models_handler_is_preset_repo(models_handler)) {
// apply the preset and start the server in router mode
common_models_handler_apply(models_handler, params);
}
} catch (const std::exception & e) {
SRV_ERR("failed to fetch model metadata: %s\n", e.what());
return 1;
}
}
// router server never loads a model and must not touch the GPU
const bool is_router_server = params.model.path.empty()
&& params.model.hf_repo.empty();
// skip device enumeration so the CUDA primary context stays uncreated
common_params_print_info(params, !is_router_server);
if (!is_router_server) {
// validate batch size for embeddings
// embeddings require all tokens to be processed in a single ubatch
// see https://github.com/ggml-org/llama.cpp/issues/12836
if (params.embedding && params.n_batch > params.n_ubatch) {
SRV_WRN("embeddings enabled with n_batch (%d) > n_ubatch (%d)\n", params.n_batch, params.n_ubatch);
SRV_WRN("setting n_batch = n_ubatch = %d to avoid assertion failure\n", params.n_ubatch);
params.n_batch = params.n_ubatch;
}
if (params.n_parallel < 0) {
SRV_TRC("%s", "n_parallel is set to auto, using n_parallel = 4 and kv_unified = true\n");
params.n_parallel = 4;
params.kv_unified = true;
}
}
// for consistency between server router mode and single-model mode, we set the same model name as alias
auto model_name = params.model.get_name();
if (params.model_alias.empty() && !model_name.empty()) {
params.model_alias.insert(model_name);
}
// note: this is guaranteed to out-live ctx_http and tools
server_mcp mcp_mgr;
// struct that contains llama context and inference
server_context ctx_server;
server_http_context ctx_http;
if (!ctx_http.init(params)) {
SRV_ERR("%s", "failed to initialize HTTP server\n");
return 1;
}
//
// Router
//
// register API routes
server_child child; // only used in non-router mode
server_routes routes(params, ctx_server);
server_tools tools;
std::optional<server_models_routes> models_routes{};
if (is_router_server) {
// setup server instances manager
try {
models_routes.emplace(params, argc, argv);
} catch (const std::exception & e) {
SRV_ERR("failed to initialize router models: %s\n", e.what());
return 1;
}
// proxy handlers
// note: routes.get_health stays the same
routes.get_metrics = models_routes->proxy_get;
routes.post_props = models_routes->proxy_post;
routes.post_completions = models_routes->proxy_post;
routes.post_completions_oai = models_routes->proxy_post;
routes.post_chat_completions = models_routes->proxy_post;
routes.post_control = models_routes->proxy_post;
routes.post_responses_oai = models_routes->proxy_post;
routes.post_transcriptions_oai = models_routes->proxy_post;
routes.post_anthropic_messages = models_routes->proxy_post;
routes.post_anthropic_count_tokens = models_routes->proxy_post;
routes.post_infill = models_routes->proxy_post;
routes.post_embeddings = models_routes->proxy_post;
routes.post_embeddings_oai = models_routes->proxy_post;
routes.post_rerank = models_routes->proxy_post;
routes.post_tokenize = models_routes->proxy_post;
routes.post_detokenize = models_routes->proxy_post;
routes.post_apply_template = models_routes->proxy_post;
routes.post_chat_completions_tok = models_routes->proxy_post;
routes.post_responses_tok_oai = models_routes->proxy_post;
routes.get_lora_adapters = models_routes->proxy_get;
routes.post_lora_adapters = models_routes->proxy_post;
routes.get_slots = models_routes->proxy_get;
routes.post_slots = models_routes->proxy_post;
// custom routes for router
routes.get_props = models_routes->get_router_props;
routes.get_models = models_routes->get_router_models;
ctx_http.post("/models", ex_wrapper(models_routes->post_router_models));
ctx_http.post("/models/load", ex_wrapper(models_routes->post_router_models_load));
ctx_http.post("/models/unload", ex_wrapper(models_routes->post_router_models_unload));
ctx_http.get ("/models/sse", ex_wrapper(models_routes->get_router_models_sse));
ctx_http.del ("/models", ex_wrapper(models_routes->del_router_models));
}
ctx_http.get ("/health", ex_wrapper(routes.get_health)); // public endpoint (no API key check)
ctx_http.get ("/v1/health", ex_wrapper(routes.get_health)); // public endpoint (no API key check)
ctx_http.get ("/metrics", ex_wrapper(routes.get_metrics));
ctx_http.get ("/props", ex_wrapper(routes.get_props));
ctx_http.post("/props", ex_wrapper(routes.post_props));
ctx_http.get ("/models", ex_wrapper(routes.get_models)); // public endpoint (no API key check)
ctx_http.get ("/v1/models", ex_wrapper(routes.get_models)); // public endpoint (no API key check)
ctx_http.post("/completion", ex_wrapper(routes.post_completions)); // legacy
ctx_http.post("/completions", ex_wrapper(routes.post_completions));
ctx_http.post("/v1/completions", ex_wrapper(routes.post_completions_oai));
ctx_http.post("/chat/completions", ex_wrapper(routes.post_chat_completions));
ctx_http.post("/v1/chat/completions", ex_wrapper(routes.post_chat_completions));
ctx_http.post("/v1/chat/completions/control", ex_wrapper(routes.post_control));
ctx_http.post("/v1/responses", ex_wrapper(routes.post_responses_oai));
ctx_http.post("/responses", ex_wrapper(routes.post_responses_oai));
ctx_http.post("/v1/audio/transcriptions", ex_wrapper(routes.post_transcriptions_oai));
ctx_http.post("/audio/transcriptions", ex_wrapper(routes.post_transcriptions_oai));
ctx_http.post("/v1/messages", ex_wrapper(routes.post_anthropic_messages)); // anthropic messages API
ctx_http.post("/infill", ex_wrapper(routes.post_infill));
ctx_http.post("/embedding", ex_wrapper(routes.post_embeddings)); // legacy
ctx_http.post("/embeddings", ex_wrapper(routes.post_embeddings));
ctx_http.post("/v1/embeddings", ex_wrapper(routes.post_embeddings_oai));
ctx_http.post("/rerank", ex_wrapper(routes.post_rerank));
ctx_http.post("/reranking", ex_wrapper(routes.post_rerank));
ctx_http.post("/v1/rerank", ex_wrapper(routes.post_rerank));
ctx_http.post("/v1/reranking", ex_wrapper(routes.post_rerank));
ctx_http.post("/tokenize", ex_wrapper(routes.post_tokenize));
ctx_http.post("/detokenize", ex_wrapper(routes.post_detokenize));
ctx_http.post("/apply-template", ex_wrapper(routes.post_apply_template));
// token counting
ctx_http.post("/chat/completions/input_tokens", ex_wrapper(routes.post_chat_completions_tok));
ctx_http.post("/v1/chat/completions/input_tokens", ex_wrapper(routes.post_chat_completions_tok));
ctx_http.post("/responses/input_tokens", ex_wrapper(routes.post_responses_tok_oai));
ctx_http.post("/v1/responses/input_tokens", ex_wrapper(routes.post_responses_tok_oai));
ctx_http.post("/v1/messages/count_tokens", ex_wrapper(routes.post_anthropic_count_tokens)); // anthropic token counting
// LoRA adapters hotswap
ctx_http.get ("/lora-adapters", ex_wrapper(routes.get_lora_adapters));
ctx_http.post("/lora-adapters", ex_wrapper(routes.post_lora_adapters));
// Save & load slots
ctx_http.get ("/slots", ex_wrapper(routes.get_slots));
ctx_http.post("/slots/:id_slot", ex_wrapper(routes.post_slots));
// resumable streaming: a child binds the local session factories, the router binds
// proxies that resolve the owning child, see server-stream.h
server_http_context::handler_t stream_get_h;
server_http_context::handler_t streams_lookup_h;
server_http_context::handler_t stream_delete_h;
if (is_router_server) {
stream_get_h = models_routes->router_stream_get;
streams_lookup_h = models_routes->router_streams_lookup;
stream_delete_h = models_routes->router_stream_delete;
} else {
stream_get_h = server_stream_make_get_handler();
streams_lookup_h = server_stream_make_lookup_handler();
stream_delete_h = server_stream_make_delete_handler();
}
ctx_http.get ("/v1/stream", ex_wrapper(stream_get_h));
ctx_http.post("/v1/streams/lookup", ex_wrapper(streams_lookup_h));
ctx_http.del ("/v1/stream", ex_wrapper(stream_delete_h));
// Google Cloud Platform (Vertex AI) compat
ctx_http.register_gcp_compat();
// return 403 for disabled features
server_http_context::handler_t res_403 = [](const server_http_req &) {
auto res = std::make_unique<server_http_res>();
res->status = 403;
res->data = safe_json_to_str({
{"error", {
{"message", "this feature is disabled"},
{"type", "feature_disabled"},
}}
});
return res;
};
if (params.cors_origins == "*" && params.api_keys.empty()) {
SRV_WRN("%s", "-----------------\n");
SRV_WRN("%s", "CORS is set to allow all origins ('*') and no API key is set\n");
SRV_WRN("%s", "this can be a security risk (cross-origin attacks)\n");
SRV_WRN("%s", "more info: https://github.com/ggml-org/llama.cpp/pull/25655\n");
SRV_WRN("%s", "-----------------\n");
}
// CORS proxy (EXPERIMENTAL, only used by the Web UI for MCP)
std::vector<std::string> warn_names;
if (is_router_server) {
warn_names.push_back("router mode");
}
if (params.ui_mcp_proxy) {
ctx_http.get ("/cors-proxy", ex_wrapper(proxy_handler_get));
ctx_http.post("/cors-proxy", ex_wrapper(proxy_handler_post));
warn_names.push_back("MCP proxy (experimental)");
} else {
ctx_http.get ("/cors-proxy", ex_wrapper(res_403));
ctx_http.post("/cors-proxy", ex_wrapper(res_403));
}
try {
mcp_mgr.start(params);
} catch (const std::exception & e) {
SRV_ERR("MCP starting failed: %s\n", e.what());
return 1;
}
if (!params.server_tools.empty() || !mcp_mgr.empty()) {
try {
tools.setup(params.server_tools, mcp_mgr, params.server_tools_runtime);
} catch (const std::exception & e) {
SRV_ERR("tools setup failed: %s\n", e.what());
return 1;
}
ctx_http.get ("/tools", ex_wrapper(tools.handle_get));
ctx_http.post("/tools", ex_wrapper(tools.handle_post));
if (!params.server_tools.empty()) {
warn_names.push_back("built-in tools (experimental)");
}
if (!params.server_tools_runtime.empty()) {
warn_names.push_back("tools runtime (experimental)");
}
if (!mcp_mgr.empty()) {
warn_names.push_back("MCP servers (experimental)");
}
} else {
ctx_http.get ("/tools", ex_wrapper(res_403));
ctx_http.post("/tools", ex_wrapper(res_403));
}
if (warn_names.size() > 0) {
SRV_WRN("%s", "-----------------\n");
SRV_WRN("%s", "the following feature(s) are enabled:\n");
for (const auto & name : warn_names) {
SRV_WRN(" %s\n", name.c_str());
}
SRV_WRN("%s", "do not expose the server to untrusted environments\n");
SRV_WRN("%s", "-----------------\n");
}
//
// Handle downloading model
//
if (child.is_child() && child.get_mode() == SERVER_CHILD_MODE_DOWNLOAD) {
return child.run_download(params);
} else if (!is_router_server && !is_run_by_cli) {
// single-model mode (NOT spawned by router)
// if this is invoked by CLI, model downloading should be already handled
try {
common_models_handler_apply(models_handler, params);
} catch (const std::exception & e) {
SRV_ERR("failed to download model: %s\n", e.what());
return 1;
}
}
//
// Start the server
//
std::function<void()> clean_up;
if (is_router_server) {
SRV_INF("%s", "starting server in router mode. models will be automatically loaded on-demand\n");
clean_up = [&models_routes, &mcp_mgr]() {
SRV_INF("%s: cleaning up before exit...\n", __func__);
// stop the session GC first, it finalizes live sessions and wakes pending readers
server_stream_session_manager_stop();
if (models_routes.has_value()) {
models_routes->stopping.store(true); // maybe redundant, but just to be safe
models_routes->models.unload_all();
}
mcp_mgr.shutdown();
llama_backend_free();
};
if (!ctx_http.start()) {
clean_up();
SRV_ERR("%s", "exiting due to HTTP server error\n");
return 1;
}
ctx_http.is_ready.store(true);
shutdown_handler = [&](int) {
if (models_routes.has_value()) {
// important to disconnect any SSE clients
models_routes->stopping.store(true);
}
mcp_mgr.shutdown();
ctx_http.stop();
};
} else {
// setup clean up function, to be called before exit
clean_up = [&ctx_http, &ctx_server, &mcp_mgr]() {
SRV_INF("%s: cleaning up before exit...\n", __func__);
// stop the session GC first, it finalizes live sessions and wakes pending readers
server_stream_session_manager_stop();
ctx_http.stop();
ctx_server.terminate();
mcp_mgr.shutdown();
llama_backend_free();
};
// start the HTTP server before loading the model to be able to serve /health requests
if (!ctx_http.start()) {
clean_up();
SRV_ERR("%s", "exiting due to HTTP server error\n");
return 1;
}
// setup communication child --> router if necessary
if (child.is_child()) {
ctx_server.set_state_callback([&](server_state state, json payload) {
child.notify_to_router(server_state_to_str(state), payload);
});
}
if (!ctx_server.load_model(params)) {
clean_up();
if (ctx_http.thread.joinable()) {
ctx_http.thread.join();
}
SRV_ERR("%s", "exiting due to model loading error\n");
return 1;
}
routes.update_meta(ctx_server);
ctx_http.is_ready.store(true);
SRV_INF("%s", "model loaded\n");
shutdown_handler = [&](int) {
mcp_mgr.shutdown();
// this will unblock start_loop()
ctx_server.terminate();
};
}
// register signal handler if not running by CLI
if (!is_run_by_cli) {
#if defined (__unix__) || (defined (__APPLE__) && defined (__MACH__))
struct sigaction sigint_action;
sigint_action.sa_handler = signal_handler;
sigemptyset (&sigint_action.sa_mask);
sigint_action.sa_flags = 0;
sigaction(SIGINT, &sigint_action, NULL);
sigaction(SIGTERM, &sigint_action, NULL);
#elif defined (_WIN32)
auto console_ctrl_handler = +[](DWORD ctrl_type) -> BOOL {
return (ctrl_type == CTRL_C_EVENT) ? (signal_handler(SIGINT), true) : false;
};
SetConsoleCtrlHandler(reinterpret_cast<PHANDLER_ROUTINE>(console_ctrl_handler), true);
#endif
}
SRV_INF("listening on %s\n", ctx_http.listening_address.c_str());
// TODO: remove this in the future
// check the string to also handle the .sock case
if (string_ends_with(ctx_http.listening_address, ":8080")) {
SRV_WRN("%s", "NOTICE: server default port will be changed to :9931 in a future release\n");
SRV_WRN("%s", " ref: https://github.com/ggml-org/llama.cpp/pull/26508\n");
}
if (is_router_server) {
if (!params.models_preset_hf.empty()) {
SRV_WRN( "NOTE: using preset.ini from HF repo '%s'\n", params.models_preset_hf.c_str());
SRV_WRN("%s", " please only use presets that you can trust! Unknown presets may be unsafe\n");
}
if (ctx_http.thread.joinable()) {
ctx_http.thread.join(); // keep the main thread alive
}
// when the HTTP server stops, clean up and exit
clean_up();
} else {
// optionally, notify router server that this instance is ready
std::thread monitor_thread;
if (child.is_child()) {
monitor_thread = child.setup(shutdown_handler);
child.notify_to_router(server_state_to_str(SERVER_STATE_READY), routes.get_model_info());
}
// this call blocks the main thread until queue_tasks.terminate() is called
ctx_server.start_loop();
clean_up();
if (ctx_http.thread.joinable()) {
ctx_http.thread.join();
}
if (monitor_thread.joinable()) {
monitor_thread.join();
}
auto * ll_ctx = ctx_server.get_llama_context();
if (ll_ctx != nullptr) {
common_memory_breakdown_print(ll_ctx);
}
}
return 0;
}
Reference in New Issue View Git Blame Copy Permalink