vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2025-09-16 20:09:41 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Merge branch 'master' into concedo_experimental

Browse source 
# Conflicts:
#	.devops/nix/package.nix
#	.github/workflows/build.yml
#	.gitignore
#	CMakeLists.txt
#	Makefile
#	README.md
#	ci/run.sh
#	flake.lock
#	flake.nix
#	scripts/get-flags.mk
#	scripts/get-wikitext-2.sh
#	scripts/sync-ggml.last
#	tests/CMakeLists.txt
#	tests/test-backend-ops.cpp
#	tests/test-grammar-parser.cpp
#	tests/test-llama-grammar.cpp
This commit is contained in:
Concedo 2024-02-20 16:30:21 +08:00
commit f0a662112b
34 changed files with 2394 additions and 753 deletions
Download patch file Download diff file Expand all files Collapse all files

75
examples/server/README.md
View file

@ -39,6 +39,8 @@ see https://github.com/ggerganov/llama.cpp/issues/1437
- `--mmproj MMPROJ_FILE`: Path to a multimodal projector file for LLaVA.
- `--grp-attn-n`: Set the group attention factor to extend context size through self-extend(default: 1=disabled), used together with group attention width `--grp-attn-w`
- `--grp-attn-w`: Set the group attention width to extend context size through self-extend(default: 512), used together with group attention factor `--grp-attn-n`
- `-n, --n-predict`: Set the maximum tokens to predict (default: -1)
- `--slots-endpoint-disable`: To disable slots state monitoring endpoint. Slots state may contain user data, prompts included.
## Build
@ -132,9 +134,11 @@ node index.js
## API Endpoints
- **GET** `/health`: Returns the current state of the server:
- `{"status": "loading model"}` if the model is still being loaded.
- `{"status": "error"}` if the model failed to load.
- `{"status": "ok"}` if the model is successfully loaded and the server is ready for further requests mentioned below.
- 503 -> `{"status": "loading model"}` if the model is still being loaded.
- 500 -> `{"status": "error"}` if the model failed to load.
- 200 -> `{"status": "ok", "slots_idle": 1, "slots_processing": 2 }` if the model is successfully loaded and the server is ready for further requests mentioned below.
- 200 -> `{"status": "no slot available", "slots_idle": 0, "slots_processing": 32}` if no slot are currently available.
- 503 -> `{"status": "no slot available", "slots_idle": 0, "slots_processing": 32}` if the query parameter `fail_on_no_slot` is provided and no slot are currently available.
- **POST** `/completion`: Given a `prompt`, it returns the predicted completion.
@ -196,6 +200,8 @@ node index.js
`n_probs`: If greater than 0, the response also contains the probabilities of top N tokens for each generated token (default: 0)
`min_keep`: If greater than 0, force samplers to return N possible tokens at minimum (default: 0)
`image_data`: An array of objects to hold base64-encoded image `data` and its `id`s to be reference in `prompt`. You can determine the place of the image in the prompt as in the following: `USER:[img-12]Describe the image in detail.\nASSISTANT:`. In this case, `[img-12]` will be replaced by the embeddings of the image with id `12` in the following `image_data` array: `{..., "image_data": [{"data": "<BASE64_STRING>", "id": 12}]}`. Use `image_data` only with multimodal models, e.g., LLaVA.
`slot_id`: Assign the completion task to an specific slot. If is -1 the task will be assigned to a Idle slot (default: -1)
@ -379,6 +385,69 @@ Notice that each `probs` is an array of length `n_probs`.
}'
```
- **GET** `/slots`: Returns the current slots processing state. Can be disabled with `--slots-endpoint-disable`.
### Result JSON
```json
[
{
"dynatemp_exponent": 1.0,
"dynatemp_range": 0.0,
"frequency_penalty": 0.0,
"grammar": "",
"id": 0,
"ignore_eos": false,
"logit_bias": [],
"min_p": 0.05000000074505806,
"mirostat": 0,
"mirostat_eta": 0.10000000149011612,
"mirostat_tau": 5.0,
"model": "llama-2-7b-32k-instruct.Q2_K.gguf",
"n_ctx": 2048,
"n_keep": 0,
"n_predict": 100000,
"n_probs": 0,
"next_token": {
"has_next_token": true,
"n_remain": -1,
"num_tokens_predicted": 0,
"stopped_eos": false,
"stopped_limit": false,
"stopped_word": false,
"stopping_word": ""
},
"penalize_nl": true,
"penalty_prompt_tokens": [],
"presence_penalty": 0.0,
"prompt": "Say hello to llama.cpp",
"repeat_last_n": 64,
"repeat_penalty": 1.100000023841858,
"samplers": [
"top_k",
"tfs_z",
"typical_p",
"top_p",
"min_p",
"temperature"
],
"seed": 42,
"state": 1,
"stop": [
"\n"
],
"stream": false,
"task_id": 0,
"temperature": 0.0,
"tfs_z": 1.0,
"top_k": 40,
"top_p": 0.949999988079071,
"typical_p": 1.0,
"use_penalty_prompt_tokens": false
}
]
```
## More examples
### Change system prompt on runtime

4
examples/server/public/index.html
View file

@ -234,6 +234,7 @@
mirostat_eta: 0.1, // learning rate
grammar: '',
n_probs: 0, // no completion_probabilities,
min_keep: 0, // min probs from each sampler,
image_data: [],
cache_prompt: true,
api_key: ''
@ -791,6 +792,9 @@
<fieldset>
${IntField({ label: "Show Probabilities", max: 10, min: 0, name: "n_probs", value: params.value.n_probs })}
</fieldset>
<fieldset>
${IntField({ label: "Min Probabilities from each Sampler", max: 10, min: 0, name: "min_keep", value: params.value.min_keep })}
</fieldset>
<fieldset>
<label for="api_key">API Key</label>
<input type="text" name="api_key" value="${params.value.api_key}" placeholder="Enter API key" oninput=${updateParams} />

107
examples/server/server.cpp
View file

@ -29,6 +29,7 @@
#include <chrono>
#include <condition_variable>
#include <atomic>
#include <signal.h>
using json = nlohmann::json;
@ -41,6 +42,7 @@ struct server_params
int32_t port = 8080;
int32_t read_timeout = 600;
int32_t write_timeout = 600;
bool slots_endpoint = true;
};
bool server_verbose = false;
@ -159,6 +161,7 @@ struct llama_client_slot
int32_t n_decoded = 0;
int32_t n_remaining = -1;
int32_t i_batch = -1;
int32_t n_predict = -1;
int32_t num_prompt_tokens = 0;
int32_t num_prompt_tokens_processed = 0;
@ -410,6 +413,7 @@ struct llama_server_context
slot.id = i;
slot.n_ctx = n_ctx_slot;
slot.n_predict = params.n_predict;
LOG_TEE(" -> Slot %i - max context: %i\n", slot.id, n_ctx_slot);
@ -545,6 +549,16 @@ struct llama_server_context
slot->params.seed = json_value(data, "seed", default_params.seed);
slot->sparams.grammar = json_value(data, "grammar", default_sparams.grammar);
slot->sparams.n_probs = json_value(data, "n_probs", default_sparams.n_probs);
slot->sparams.min_keep = json_value(data, "min_keep", default_sparams.min_keep);
if (slot->n_predict > 0 && slot->params.n_predict > slot->n_predict) {
// Might be better to reject the request with a 400 ?
LOG_WARNING("Max tokens to predict exceeds server configuration", {
{"params.n_predict", slot->params.n_predict},
{"slot.n_predict", slot->n_predict},
});
slot->params.n_predict = slot->n_predict;
}
// infill
if (data.count("input_prefix") != 0)
@ -1053,6 +1067,7 @@ struct llama_server_context
return json {
{"n_ctx", slot.n_ctx},
{"n_predict", slot.n_predict},
{"model", params.model_alias},
{"seed", slot.params.seed},
{"temperature", slot.sparams.temp},
@ -1080,6 +1095,7 @@ struct llama_server_context
{"stream", slot.params.stream},
{"logit_bias", slot.sparams.logit_bias},
{"n_probs", slot.sparams.n_probs},
{"min_keep", slot.sparams.min_keep},
{"grammar", slot.sparams.grammar},
{"samplers", samplers_sequence}
};
@ -1914,14 +1930,16 @@ static void server_print_usage(const char *argv0, const gpt_params &params,
printf(" set a file to load a system prompt (initial prompt of all slots), this is useful for chat applications.\n");
printf(" --mmproj MMPROJ_FILE path to a multimodal projector file for LLaVA.\n");
printf(" --log-disable disables logging to a file.\n");
printf(" --slots-endpoint-disable disables slots monitoring endpoint.\n");
printf("\n");
printf(" -n, --n-predict maximum tokens to predict (default: %d)\n", params.n_predict);
printf(" --override-kv KEY=TYPE:VALUE\n");
printf(" advanced option to override model metadata by key. may be specified multiple times.\n");
printf(" types: int, float, bool. example: --override-kv tokenizer.ggml.add_bos_token=bool:false\n");
printf(" -gan N, --grp-attn-n N set the group attention factor to extend context size through self-extend(default: 1=disabled), used together with group attention width `--grp-attn-w`");
printf(" -gaw N, --grp-attn-w N set the group attention width to extend context size through self-extend(default: 512), used together with group attention factor `--grp-attn-n`");
printf(" --chat-template FORMAT_NAME");
printf(" set chat template, possible valus is: llama2, chatml (default %s)", sparams.chat_template.c_str());
printf(" set chat template, possible value is: llama2, chatml (default %s)", sparams.chat_template.c_str());
printf("\n");
}
@ -2361,6 +2379,10 @@ static void server_params_parse(int argc, char **argv, server_params &sparams,
log_set_target(stdout);
LOG_INFO("logging to file is disabled.", {});
}
else if (arg == "--slots-endpoint-disable")
{
sparams.slots_endpoint = false;
}
else if (arg == "--chat-template")
{
if (++i >= argc)
@ -2512,6 +2534,9 @@ static void append_to_generated_text_from_generated_token_probs(llama_server_con
}
}
std::function<void(int)> shutdown_handler;
inline void signal_handler(int signal) { shutdown_handler(signal); }
int main(int argc, char **argv)
{
#if SERVER_VERBOSE != 1
@ -2558,13 +2583,40 @@ int main(int argc, char **argv)
res.set_header("Access-Control-Allow-Headers", "*");
});
svr.Get("/health", [&](const httplib::Request&, httplib::Response& res) {
svr.Get("/health", [&](const httplib::Request& req, httplib::Response& res) {
server_state current_state = state.load();
switch(current_state) {
case SERVER_STATE_READY:
res.set_content(R"({"status": "ok"})", "application/json");
res.status = 200; // HTTP OK
case SERVER_STATE_READY: {
int available_slots = 0;
int processing_slots = 0;
for (llama_client_slot &slot: llama.slots) {
if (slot.available()) {
available_slots++;
} else {
processing_slots++;
}
}
if (available_slots > 0) {
json health = {
{"status", "ok"},
{"slots_idle", available_slots},
{"slots_processing", processing_slots}};
res.set_content(health.dump(), "application/json");
res.status = 200; // HTTP OK
} else {
json health = {
{"status", "no slot available"},
{"slots_idle", available_slots},
{"slots_processing", processing_slots}};
res.set_content(health.dump(), "application/json");
if (req.has_param("fail_on_no_slot")) {
res.status = 503; // HTTP Service Unavailable
} else {
res.status = 200; // HTTP OK
}
}
break;
}
case SERVER_STATE_LOADING_MODEL:
res.set_content(R"({"status": "loading model"})", "application/json");
res.status = 503; // HTTP Service Unavailable
@ -2576,6 +2628,32 @@ int main(int argc, char **argv)
}
});
if (sparams.slots_endpoint) {
svr.Get("/slots", [&](const httplib::Request&, httplib::Response& res) {
json slots;
for (llama_client_slot & slot : llama.slots) {
json slot_data = llama.get_formated_generation(slot);
slot_data["id"] = slot.id;
slot_data["task_id"] = slot.task_id;
slot_data["state"] = slot.state;
slot_data["prompt"] = slot.prompt;
slot_data["next_token"] = {
{"has_next_token", slot.has_next_token},
{"n_remain", slot.n_remaining},
{"num_tokens_predicted", slot.n_decoded},
{"stopped_eos", slot.stopped_eos},
{"stopped_word", slot.stopped_word},
{"stopped_limit", slot.stopped_limit},
{"stopping_word", slot.stopping_word},
};
slots.push_back(slot_data);
}
res.set_content(slots.dump(), "application/json");
res.status = 200; // HTTP OK
});
}
svr.set_logger(log_server_request);
svr.set_exception_handler([](const httplib::Request &, httplib::Response &res, std::exception_ptr ep)
@ -3129,8 +3207,25 @@ int main(int argc, char **argv)
std::placeholders::_2,
std::placeholders::_3
));
llama.queue_tasks.start_loop();
shutdown_handler = [&](int) {
llama.queue_tasks.terminate();
};
#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);
#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
llama.queue_tasks.start_loop();
svr.stop();
t.join();
llama_backend_free();

20
examples/server/utils.hpp
View file

@ -220,6 +220,7 @@ inline std::string format_chatml(std::vector<json> messages)
struct llama_server_queue {
int id = 0;
std::mutex mutex_tasks;
bool running;
// queues
std::vector<task_server> queue_tasks;
std::vector<task_server> queue_tasks_deferred;
@ -278,9 +279,18 @@ struct llama_server_queue {
queue_tasks_deferred.clear();
}
// Start the main loop. This call is blocking
[[noreturn]]
// end the start_loop routine
void terminate() {
{
std::unique_lock<std::mutex> lock(mutex_tasks);
running = false;
}
condition_tasks.notify_all();
}
// Start the main loop.
void start_loop() {
running = true;
while (true) {
// new task arrived
LOG_VERBOSE("have new task", {});
@ -324,8 +334,12 @@ struct llama_server_queue {
{
std::unique_lock<std::mutex> lock(mutex_tasks);
if (queue_tasks.empty()) {
if (!running) {
LOG_VERBOSE("ending start_loop", {});
return;
}
condition_tasks.wait(lock, [&]{
return !queue_tasks.empty();
return (!queue_tasks.empty() || !running);
});
}
}