common : move arg parser code to arg.cpp (#9388)

* common : move arg parser to arg.cpp

* better categorize args

* add cmake

* missing climits

* missing cstdarg

* common : more explicit includes

* fix build

* refactor gpt_params_parse

* update server readme

* fix test

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>

Makefile

@@ -925,6 +925,7 @@ OBJ_LLAMA = \
 
 OBJ_COMMON = \
 	common/common.o \
+	common/arg.o \
 	common/console.o \
 	common/ngram-cache.o \
 	common/sampling.o \
@@ -1157,6 +1158,11 @@ common/common.o: \
 	include/llama.h
 	$(CXX) $(CXXFLAGS) -c $< -o $@
 
+common/arg.o: \
+	common/arg.cpp \
+	common/arg.h
+	$(CXX) $(CXXFLAGS) -c $< -o $@
+
 common/sampling.o: \
 	common/sampling.cpp \
 	common/sampling.h \

common/CMakeLists.txt

@@ -54,6 +54,8 @@ add_library(${TARGET} STATIC
     base64.hpp
     common.h
     common.cpp
+    arg.h
+    arg.cpp
     sampling.h
     sampling.cpp
     console.h

common/arg.h

@@ -0,0 +1,77 @@
+#pragma once
+
+#include "common.h"
+
+#include <set>
+#include <string>
+#include <vector>
+
+//
+// CLI argument parsing
+//
+
+struct llama_arg {
+    std::set<enum llama_example> examples = {LLAMA_EXAMPLE_COMMON};
+    std::vector<const char *> args;
+    const char * value_hint   = nullptr; // help text or example for arg value
+    const char * value_hint_2 = nullptr; // for second arg value
+    const char * env          = nullptr;
+    std::string help;
+    bool is_sparam = false; // is current arg a sampling param?
+    void (*handler_void)   (gpt_params & params) = nullptr;
+    void (*handler_string) (gpt_params & params, const std::string &) = nullptr;
+    void (*handler_str_str)(gpt_params & params, const std::string &, const std::string &) = nullptr;
+    void (*handler_int)    (gpt_params & params, int) = nullptr;
+
+    llama_arg(
+        const std::initializer_list<const char *> & args,
+        const char * value_hint,
+        const std::string & help,
+        void (*handler)(gpt_params & params, const std::string &)
+    ) : args(args), value_hint(value_hint), help(help), handler_string(handler) {}
+
+    llama_arg(
+        const std::initializer_list<const char *> & args,
+        const char * value_hint,
+        const std::string & help,
+        void (*handler)(gpt_params & params, int)
+    ) : args(args), value_hint(value_hint), help(help), handler_int(handler) {}
+
+    llama_arg(
+        const std::initializer_list<const char *> & args,
+        const std::string & help,
+        void (*handler)(gpt_params & params)
+    ) : args(args), help(help), handler_void(handler) {}
+
+    // support 2 values for arg
+    llama_arg(
+        const std::initializer_list<const char *> & args,
+        const char * value_hint,
+        const char * value_hint_2,
+        const std::string & help,
+        void (*handler)(gpt_params & params, const std::string &, const std::string &)
+    ) : args(args), value_hint(value_hint), value_hint_2(value_hint_2), help(help), handler_str_str(handler) {}
+
+    llama_arg & set_examples(std::initializer_list<enum llama_example> examples);
+    llama_arg & set_env(const char * env);
+    llama_arg & set_sparam();
+    bool in_example(enum llama_example ex);
+    bool get_value_from_env(std::string & output);
+    bool has_value_from_env();
+    std::string to_string();
+};
+
+struct gpt_params_context {
+    enum llama_example ex = LLAMA_EXAMPLE_COMMON;
+    gpt_params & params;
+    std::vector<llama_arg> options;
+    void(*print_usage)(int, char **) = nullptr;
+    gpt_params_context(gpt_params & params) : params(params) {}
+};
+
+// parse input arguments from CLI
+// if one argument has invalid value, it will automatically display usage of the specific argument (and not the full usage message)
+bool gpt_params_parse(int argc, char ** argv, gpt_params & params, llama_example ex, void(*print_usage)(int, char **) = nullptr);
+
+// function to be used by test-arg-parser
+gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex, void(*print_usage)(int, char **) = nullptr);

common/common.h

@@ -4,20 +4,11 @@
 
 #include "llama.h"
 
-#include "sampling.h"
-
 #define LOG_NO_FILE_LINE_FUNCTION
 #include "log.h"
 
-#include <cmath>
 #include <string>
 #include <vector>
-#include <random>
-#include <thread>
-#include <set>
-#include <unordered_map>
-#include <tuple>
-#include <functional>
 
 #ifdef _WIN32
 #define DIRECTORY_SEPARATOR '\\'
@@ -56,11 +47,20 @@ struct llama_control_vector_load_info;
 // CPU utils
 //
 
+struct cpu_params {
+    int      n_threads                   = -1;
+    bool     cpumask[GGML_MAX_N_THREADS] = {false}; // CPU affinity mask.
+    bool     mask_valid                  = false;   // Default: any CPU
+    enum ggml_sched_priority  priority   = GGML_SCHED_PRIO_NORMAL;  // Scheduling prio : (0 - normal, 1 - medium, 2 - high, 3 - realtime)
+    bool     strict_cpu                  = false;   // Use strict CPU placement
+    uint32_t poll                        = 50;      // Polling (busywait) level (0 - no polling, 100 - mostly polling)
+};
+
 int32_t cpu_get_num_physical_cores();
 int32_t cpu_get_num_math();
 
 //
-// CLI argument parsing
+// Common params
 //
 
 enum llama_example {
@@ -78,28 +78,71 @@ enum llama_example {
     LLAMA_EXAMPLE_CVECTOR_GENERATOR,
     LLAMA_EXAMPLE_EXPORT_LORA,
     LLAMA_EXAMPLE_LLAVA,
+    LLAMA_EXAMPLE_LOOKUP,
+    LLAMA_EXAMPLE_PARALLEL,
 
     LLAMA_EXAMPLE_COUNT,
 };
 
+enum gpt_sampler_type {
+    GPT_SAMPLER_TYPE_NONE        = 0,
+    GPT_SAMPLER_TYPE_TOP_K       = 1,
+    GPT_SAMPLER_TYPE_TOP_P       = 2,
+    GPT_SAMPLER_TYPE_MIN_P       = 3,
+    GPT_SAMPLER_TYPE_TFS_Z       = 4,
+    GPT_SAMPLER_TYPE_TYPICAL_P   = 5,
+    GPT_SAMPLER_TYPE_TEMPERATURE = 6,
+};
+
 // dimensionality reduction methods, used by cvector-generator
 enum dimre_method {
     DIMRE_METHOD_PCA,
     DIMRE_METHOD_MEAN,
 };
 
-struct cpu_params {
+// sampler parameters
-    int      n_threads                   = -1;
+struct gpt_sampler_params {
-    bool     cpumask[GGML_MAX_N_THREADS] = {false}; // CPU affinity mask.
+    uint32_t seed = LLAMA_DEFAULT_SEED; // the seed used to initialize llama_sampler
-    bool     mask_valid                  = false;   // Default: any CPU
+
-    enum ggml_sched_priority  priority   = GGML_SCHED_PRIO_NORMAL;  // Scheduling prio : (0 - normal, 1 - medium, 2 - high, 3 - realtime)
+    int32_t n_prev            = 64;    // number of previous tokens to remember
-    bool     strict_cpu                  = false;   // Use strict CPU placement
+    int32_t n_probs           = 0;     // if greater than 0, output the probabilities of top n_probs tokens.
-    uint32_t poll                        = 50;      // Polling (busywait) level (0 - no polling, 100 - mostly polling)
+    int32_t min_keep          = 0;     // 0 = disabled, otherwise samplers should return at least min_keep tokens
+    int32_t top_k             = 40;    // <= 0 to use vocab size
+    float   top_p             = 0.95f; // 1.0 = disabled
+    float   min_p             = 0.05f; // 0.0 = disabled
+    float   tfs_z             = 1.00f; // 1.0 = disabled
+    float   typ_p             = 1.00f; // typical_p, 1.0 = disabled
+    float   temp              = 0.80f; // <= 0.0 to sample greedily, 0.0 to not output probabilities
+    float   dynatemp_range    = 0.00f; // 0.0 = disabled
+    float   dynatemp_exponent = 1.00f; // controls how entropy maps to temperature in dynamic temperature sampler
+    int32_t penalty_last_n    = 64;    // last n tokens to penalize (0 = disable penalty, -1 = context size)
+    float   penalty_repeat    = 1.00f; // 1.0 = disabled
+    float   penalty_freq      = 0.00f; // 0.0 = disabled
+    float   penalty_present   = 0.00f; // 0.0 = disabled
+    int32_t mirostat          = 0;     // 0 = disabled, 1 = mirostat, 2 = mirostat 2.0
+    float   mirostat_tau      = 5.00f; // target entropy
+    float   mirostat_eta      = 0.10f; // learning rate
+    bool    penalize_nl       = false; // consider newlines as a repeatable token
+    bool    ignore_eos        = false;
+
+    std::vector<enum gpt_sampler_type> samplers = {
+        GPT_SAMPLER_TYPE_TOP_K,
+        GPT_SAMPLER_TYPE_TFS_Z,
+        GPT_SAMPLER_TYPE_TYPICAL_P,
+        GPT_SAMPLER_TYPE_TOP_P,
+        GPT_SAMPLER_TYPE_MIN_P,
+        GPT_SAMPLER_TYPE_TEMPERATURE
+    };
+
+    std::string grammar; // optional BNF-like grammar to constrain sampling
+
+    std::vector<llama_logit_bias> logit_bias; // logit biases to apply
+
+    // print the parameters into a string
+    std::string print() const;
 };
 
 struct gpt_params {
-    enum llama_example curr_ex    = LLAMA_EXAMPLE_COMMON;
-
     int32_t n_predict             =    -1; // new tokens to predict
     int32_t n_ctx                 =     0; // context size
     int32_t n_batch               =  2048; // logical batch size for prompt processing (must be >=32 to use BLAS)
@@ -143,23 +186,23 @@ struct gpt_params {
 
     struct gpt_sampler_params sparams;
 
-    std::string model                = ""; // model path
+    std::string model                = ""; // model path                                                    // NOLINT
-    std::string model_draft          = ""; // draft model for speculative decoding
+    std::string model_draft          = ""; // draft model for speculative decoding                          // NOLINT
-    std::string model_alias          = "unknown"; // model alias
+    std::string model_alias          = "unknown"; // model alias                                            // NOLINT
-    std::string model_url            = ""; // model url to download
+    std::string model_url            = ""; // model url to download                                         // NOLINT
-    std::string hf_token             = ""; // HF token
+    std::string hf_token             = ""; // HF token                                                      // NOLINT
-    std::string hf_repo              = ""; // HF repo
+    std::string hf_repo              = ""; // HF repo                                                       // NOLINT
-    std::string hf_file              = ""; // HF file
+    std::string hf_file              = ""; // HF file                                                       // NOLINT
-    std::string prompt               = "";
+    std::string prompt               = "";                                                                  // NOLINT
-    std::string prompt_file          = ""; // store the external prompt file name
+    std::string prompt_file          = ""; // store the external prompt file name                           // NOLINT
-    std::string path_prompt_cache    = ""; // path to file for saving/loading prompt eval state
+    std::string path_prompt_cache    = ""; // path to file for saving/loading prompt eval state             // NOLINT
-    std::string input_prefix         = ""; // string to prefix user inputs with
+    std::string input_prefix         = ""; // string to prefix user inputs with                             // NOLINT
-    std::string input_suffix         = ""; // string to suffix user inputs with
+    std::string input_suffix         = ""; // string to suffix user inputs with                             // NOLINT
-    std::string logdir               = ""; // directory in which to save YAML log files
+    std::string logdir               = ""; // directory in which to save YAML log files                     // NOLINT
-    std::string lookup_cache_static  = ""; // path of static ngram cache file for lookup decoding
+    std::string lookup_cache_static  = ""; // path of static ngram cache file for lookup decoding           // NOLINT
-    std::string lookup_cache_dynamic = ""; // path of dynamic ngram cache file for lookup decoding
+    std::string lookup_cache_dynamic = ""; // path of dynamic ngram cache file for lookup decoding          // NOLINT
-    std::string logits_file          = ""; // file for saving *all* logits
+    std::string logits_file          = ""; // file for saving *all* logits                                  // NOLINT
-    std::string rpc_servers          = ""; // comma separated list of RPC servers
+    std::string rpc_servers          = ""; // comma separated list of RPC servers                           // NOLINT
 
     std::vector<std::string> in_files;   // all input files
     std::vector<std::string> antiprompt; // strings upon which more user input is prompted (a.k.a. reverse prompts)
@@ -189,7 +232,6 @@ struct gpt_params {
 
     bool   kl_divergence    = false; // compute KL divergence
 
-    std::function<void(int, char **)> print_usage = nullptr; // print example-specific usage and example
     bool usage             = false; // print usage
     bool use_color         = false; // use color to distinguish generations and inputs
     bool special           = false; // enable special token output
@@ -220,7 +262,7 @@ struct gpt_params {
     std::string cache_type_v = "f16"; // KV cache data type for the V
 
     // multimodal models (see examples/llava)
-    std::string mmproj = "";        // path to multimodal projector
+    std::string mmproj = "";        // path to multimodal projector                                         // NOLINT
     std::vector<std::string> image; // path to image file(s)
 
     // embedding
@@ -236,15 +278,15 @@ struct gpt_params {
     int     n_threads_http = -1;           // number of threads to process HTTP requests (TODO: support threadpool)
 
     std::string hostname      = "127.0.0.1";
-    std::string public_path   = "";
+    std::string public_path   = "";                                                                         // NOLINT
-    std::string chat_template = "";
+    std::string chat_template = "";                                                                         // NOLINT
-    std::string system_prompt = "";
+    std::string system_prompt = "";                                                                         // NOLINT
     bool enable_chat_template = true;
 
     std::vector<std::string> api_keys;
 
-    std::string ssl_file_key  = "";
+    std::string ssl_file_key  = "";                                                                         // NOLINT
-    std::string ssl_file_cert = "";
+    std::string ssl_file_cert = "";                                                                         // NOLINT
 
     bool endpoint_slots   = true;
     bool endpoint_metrics = false;
@@ -299,92 +341,6 @@ struct gpt_params {
     bool batched_bench_output_jsonl = false;
 };
 
-struct llama_arg {
-    std::set<enum llama_example> examples = {LLAMA_EXAMPLE_COMMON};
-    std::vector<const char *> args;
-    const char * value_hint   = nullptr; // help text or example for arg value
-    const char * value_hint_2 = nullptr; // for second arg value
-    const char * env          = nullptr;
-    std::string help;
-    void (*handler_void)   (gpt_params & params) = nullptr;
-    void (*handler_string) (gpt_params & params, const std::string &) = nullptr;
-    void (*handler_str_str)(gpt_params & params, const std::string &, const std::string &) = nullptr;
-    void (*handler_int)    (gpt_params & params, int) = nullptr;
-
-    llama_arg(
-        const std::initializer_list<const char *> & args,
-        const char * value_hint,
-        const std::string & help,
-        void (*handler)(gpt_params & params, const std::string &)
-    ) : args(args), value_hint(value_hint), help(help), handler_string(handler) {}
-
-    llama_arg(
-        const std::initializer_list<const char *> & args,
-        const char * value_hint,
-        const std::string & help,
-        void (*handler)(gpt_params & params, int)
-    ) : args(args), value_hint(value_hint), help(help), handler_int(handler) {}
-
-    llama_arg(
-        const std::initializer_list<const char *> & args,
-        const std::string & help,
-        void (*handler)(gpt_params & params)
-    ) : args(args), help(help), handler_void(handler) {}
-
-    // support 2 values for arg
-    llama_arg(
-        const std::initializer_list<const char *> & args,
-        const char * value_hint,
-        const char * value_hint_2,
-        const std::string & help,
-        void (*handler)(gpt_params & params, const std::string &, const std::string &)
-    ) : args(args), value_hint(value_hint), value_hint_2(value_hint_2), help(help), handler_str_str(handler) {}
-
-    llama_arg & set_examples(std::initializer_list<enum llama_example> examples) {
-        this->examples = std::move(examples);
-        return *this;
-    }
-
-    llama_arg & set_env(const char * env) {
-        help = help + "\n(env: " + env + ")";
-        this->env = env;
-        return *this;
-    }
-
-    bool in_example(enum llama_example ex) {
-        return examples.find(ex) != examples.end();
-    }
-
-    bool get_value_from_env(std::string & output) const {
-        if (env == nullptr) return false;
-        char * value = std::getenv(env);
-        if (value) {
-            output = value;
-            return true;
-        }
-        return false;
-    }
-
-    bool has_value_from_env() const {
-        return env != nullptr && std::getenv(env);
-    }
-
-    std::string to_string();
-};
-
-// initialize list of options (arguments) that can be used by the current example
-std::vector<llama_arg> gpt_params_parser_init(gpt_params & params, llama_example ex);
-// optionally, we can provide "print_usage" to print example usage
-std::vector<llama_arg> gpt_params_parser_init(gpt_params & params, llama_example ex, std::function<void(int, char **)> print_usage);
-
-// parse input arguments from CLI
-// if one argument has invalid value, it will automatically display usage of the specific argument (and not the full usage message)
-bool gpt_params_parse   (int argc, char ** argv, gpt_params & params, std::vector<llama_arg> & options);
-bool gpt_params_parse_ex(int argc, char ** argv, gpt_params & params, std::vector<llama_arg> & options);
-
-// print full usage message; it will be called internally by gpt_params_parse() if "-h" is set
-void gpt_params_print_usage(gpt_params & params, std::vector<llama_arg> & options);
-
 std::string gpt_params_get_system_info(const gpt_params & params);
 
 bool parse_cpu_range(const std::string& range, bool(&boolmask)[GGML_MAX_N_THREADS]);

common/sampling.cpp

@@ -2,6 +2,9 @@
 
 #include "common.h"
 
+#include <cmath>
+#include <unordered_map>
+
 // the ring buffer works similarly to std::deque, but with a fixed capacity
 // TODO: deduplicate with llama-impl.h
 template<typename T>
@@ -420,7 +423,7 @@ std::vector<gpt_sampler_type> gpt_sampler_types_from_names(const std::vector<std
 }
 
 std::vector<gpt_sampler_type> gpt_sampler_types_from_chars(const std::string & chars) {
-    std::unordered_map<char, gpt_sampler_type> sampler_name_map {
+    std::unordered_map<char, gpt_sampler_type> sampler_name_map = {
         { gpt_sampler_type_to_chr(GPT_SAMPLER_TYPE_TOP_K),       GPT_SAMPLER_TYPE_TOP_K },
         { gpt_sampler_type_to_chr(GPT_SAMPLER_TYPE_TFS_Z),       GPT_SAMPLER_TYPE_TFS_Z },
         { gpt_sampler_type_to_chr(GPT_SAMPLER_TYPE_TYPICAL_P),   GPT_SAMPLER_TYPE_TYPICAL_P },

common/sampling.h

@@ -2,61 +2,11 @@
 
 #include "llama.h"
 
+#include "common.h"
+
 #include <string>
 #include <vector>
 
-enum gpt_sampler_type {
-    GPT_SAMPLER_TYPE_NONE        = 0,
-    GPT_SAMPLER_TYPE_TOP_K       = 1,
-    GPT_SAMPLER_TYPE_TOP_P       = 2,
-    GPT_SAMPLER_TYPE_MIN_P       = 3,
-    GPT_SAMPLER_TYPE_TFS_Z       = 4,
-    GPT_SAMPLER_TYPE_TYPICAL_P   = 5,
-    GPT_SAMPLER_TYPE_TEMPERATURE = 6,
-};
-
-// sampling parameters
-struct gpt_sampler_params {
-    uint32_t seed = LLAMA_DEFAULT_SEED; // the seed used to initialize llama_sampler
-
-    int32_t n_prev            = 64;    // number of previous tokens to remember
-    int32_t n_probs           = 0;     // if greater than 0, output the probabilities of top n_probs tokens.
-    int32_t min_keep          = 0;     // 0 = disabled, otherwise samplers should return at least min_keep tokens
-    int32_t top_k             = 40;    // <= 0 to use vocab size
-    float   top_p             = 0.95f; // 1.0 = disabled
-    float   min_p             = 0.05f; // 0.0 = disabled
-    float   tfs_z             = 1.00f; // 1.0 = disabled
-    float   typ_p             = 1.00f; // typical_p, 1.0 = disabled
-    float   temp              = 0.80f; // <= 0.0 to sample greedily, 0.0 to not output probabilities
-    float   dynatemp_range    = 0.00f; // 0.0 = disabled
-    float   dynatemp_exponent = 1.00f; // controls how entropy maps to temperature in dynamic temperature sampler
-    int32_t penalty_last_n    = 64;    // last n tokens to penalize (0 = disable penalty, -1 = context size)
-    float   penalty_repeat    = 1.00f; // 1.0 = disabled
-    float   penalty_freq      = 0.00f; // 0.0 = disabled
-    float   penalty_present   = 0.00f; // 0.0 = disabled
-    int32_t mirostat          = 0;     // 0 = disabled, 1 = mirostat, 2 = mirostat 2.0
-    float   mirostat_tau      = 5.00f; // target entropy
-    float   mirostat_eta      = 0.10f; // learning rate
-    bool    penalize_nl       = false; // consider newlines as a repeatable token
-    bool    ignore_eos        = false;
-
-    std::vector<enum gpt_sampler_type> samplers = {
-        GPT_SAMPLER_TYPE_TOP_K,
-        GPT_SAMPLER_TYPE_TFS_Z,
-        GPT_SAMPLER_TYPE_TYPICAL_P,
-        GPT_SAMPLER_TYPE_TOP_P,
-        GPT_SAMPLER_TYPE_MIN_P,
-        GPT_SAMPLER_TYPE_TEMPERATURE
-    };
-
-    std::string grammar; // optional BNF-like grammar to constrain sampling
-
-    std::vector<llama_logit_bias> logit_bias; // logit biases to apply
-
-    // print the parameters into a string
-    std::string print() const;
-};
-
 // gpt_sampler extends llama_sampler with additional functionality:
 //
 //  - grammar support

examples/batched-bench/batched-bench.cpp

@@ -1,3 +1,4 @@
+#include "arg.h"
 #include "common.h"
 #include "llama.h"
 
@@ -37,8 +38,7 @@ static void print_usage(int, char ** argv) {
 int main(int argc, char ** argv) {
     gpt_params params;
 
-    auto options = gpt_params_parser_init(params, LLAMA_EXAMPLE_BENCH, print_usage);
+    if (!gpt_params_parse(argc, argv, params, LLAMA_EXAMPLE_BENCH, print_usage)) {
-    if (!gpt_params_parse(argc, argv, params, options)) {
         return 1;
     }
 

examples/batched/batched.cpp

@@ -1,3 +1,4 @@
+#include "arg.h"
 #include "common.h"
 #include "llama.h"
 
@@ -18,8 +19,7 @@ int main(int argc, char ** argv) {
     params.prompt = "Hello my name is";
     params.n_predict = 32;
 
-    auto options = gpt_params_parser_init(params, LLAMA_EXAMPLE_COMMON, print_usage);
+    if (!gpt_params_parse(argc, argv, params, LLAMA_EXAMPLE_COMMON, print_usage)) {
-    if (!gpt_params_parse(argc, argv, params, options)) {
         return 1;
     }
 

examples/cvector-generator/cvector-generator.cpp

@@ -1,3 +1,4 @@
+#include "arg.h"
 #include "common.h"
 #include "llama.h"
 #include "ggml.h"
@@ -388,8 +389,7 @@ static int prepare_entries(gpt_params & params, train_context & ctx_train) {
 int main(int argc, char ** argv) {
     gpt_params params;
 
-    auto options = gpt_params_parser_init(params, LLAMA_EXAMPLE_CVECTOR_GENERATOR, print_usage);
+    if (!gpt_params_parse(argc, argv, params, LLAMA_EXAMPLE_CVECTOR_GENERATOR, print_usage)) {
-    if (!gpt_params_parse(argc, argv, params, options)) {
         return 1;
     }
 

examples/cvector-generator/pca.hpp

@@ -12,12 +12,9 @@
 
 #include <cstdio>
 #include <ctime>
+#include <random>
 #include <string>
-#include <tuple>
 #include <vector>
-#include <algorithm>
-#include <iostream>
-#include <fstream>
 
 #define DEBUG_POS 5
 

examples/embedding/embedding.cpp

@@ -1,3 +1,4 @@
+#include "arg.h"
 #include "common.h"
 #include "llama.h"
 
@@ -79,8 +80,7 @@ static void batch_decode(llama_context * ctx, llama_batch & batch, float * outpu
 int main(int argc, char ** argv) {
     gpt_params params;
 
-    auto options = gpt_params_parser_init(params, LLAMA_EXAMPLE_EMBEDDING);
+    if (!gpt_params_parse(argc, argv, params, LLAMA_EXAMPLE_EMBEDDING)) {
-    if (!gpt_params_parse(argc, argv, params, options)) {
         return 1;
     }
 

examples/eval-callback/eval-callback.cpp

@@ -1,3 +1,4 @@
+#include "arg.h"
 #include "common.h"
 #include "llama.h"
 #include "ggml.h"
@@ -144,8 +145,7 @@ int main(int argc, char ** argv) {
 
     gpt_params params;
 
-    auto options = gpt_params_parser_init(params, LLAMA_EXAMPLE_COMMON);
+    if (!gpt_params_parse(argc, argv, params, LLAMA_EXAMPLE_COMMON)) {
-    if (!gpt_params_parse(argc, argv, params, options)) {
         return 1;
     }
 

examples/export-lora/export-lora.cpp

@@ -1,3 +1,4 @@
+#include "arg.h"
 #include "common.h"
 #include "ggml.h"
 #include "ggml-alloc.h"
@@ -401,8 +402,7 @@ static void print_usage(int, char ** argv) {
 int main(int argc, char ** argv) {
     gpt_params params;
 
-    auto options = gpt_params_parser_init(params, LLAMA_EXAMPLE_EXPORT_LORA, print_usage);
+    if (!gpt_params_parse(argc, argv, params, LLAMA_EXAMPLE_EXPORT_LORA, print_usage)) {
-    if (!gpt_params_parse(argc, argv, params, options)) {
         return 1;
     }
 

examples/gen-docs/gen-docs.cpp

@@ -1,3 +1,4 @@
+#include "arg.h"
 #include "common.h"
 
 #include <fstream>
@@ -9,11 +10,11 @@ static void export_md(std::string fname, llama_example ex) {
     std::ofstream file(fname, std::ofstream::out | std::ofstream::trunc);
 
     gpt_params params;
-    auto options = gpt_params_parser_init(params, ex);
+    auto ctx_arg = gpt_params_parser_init(params, ex);
 
     file << "| Argument | Explanation |\n";
     file << "| -------- | ----------- |\n";
-    for (auto & opt : options) {
+    for (auto & opt : ctx_arg.options) {
         file << "| `";
         // args
         for (const auto & arg : opt.args) {

examples/gritlm/gritlm.cpp

@@ -1,3 +1,4 @@
+#include "arg.h"
 #include "common.h"
 #include "llama.h"
 
@@ -153,8 +154,7 @@ static std::string gritlm_instruction(const std::string & instruction) {
 int main(int argc, char * argv[]) {
     gpt_params params;
 
-    auto options = gpt_params_parser_init(params, LLAMA_EXAMPLE_COMMON);
+    if (!gpt_params_parse(argc, argv, params, LLAMA_EXAMPLE_COMMON)) {
-    if (!gpt_params_parse(argc, argv, params, options)) {
         return 1;
     }
 

examples/imatrix/imatrix.cpp

@@ -1,3 +1,4 @@
+#include "arg.h"
 #include "common.h"
 #include "llama.h"
 
@@ -577,8 +578,7 @@ int main(int argc, char ** argv) {
     params.logits_all = true;
     params.verbosity = 1;
 
-    auto options = gpt_params_parser_init(params, LLAMA_EXAMPLE_IMATRIX, print_usage);
+    if (!gpt_params_parse(argc, argv, params, LLAMA_EXAMPLE_IMATRIX, print_usage)) {
-    if (!gpt_params_parse(argc, argv, params, options)) {
         return 1;
     }
 

examples/infill/infill.cpp

@@ -1,6 +1,7 @@
+#include "arg.h"
 #include "common.h"
-
 #include "console.h"
+#include "sampling.h"
 #include "llama.h"
 
 #include <cassert>
@@ -105,8 +106,7 @@ int main(int argc, char ** argv) {
     gpt_params params;
     g_params = &params;
 
-    auto options = gpt_params_parser_init(params, LLAMA_EXAMPLE_INFILL);
+    if (!gpt_params_parse(argc, argv, params, LLAMA_EXAMPLE_INFILL)) {
-    if (!gpt_params_parse(argc, argv, params, options)) {
         return 1;
     }
 

examples/llava/llava-cli.cpp

@@ -1,11 +1,12 @@
-#include "ggml.h"
+#include "arg.h"
+#include "base64.hpp"
 #include "log.h"
 #include "common.h"
+#include "sampling.h"
 #include "clip.h"
 #include "llava.h"
 #include "llama.h"
-
+#include "ggml.h"
-#include "base64.hpp"
 
 #include <cstdio>
 #include <cstdlib>
@@ -278,8 +279,7 @@ int main(int argc, char ** argv) {
 
     gpt_params params;
 
-    auto options = gpt_params_parser_init(params, LLAMA_EXAMPLE_LLAVA, print_usage);
+    if (!gpt_params_parse(argc, argv, params, LLAMA_EXAMPLE_LLAVA, print_usage)) {
-    if (!gpt_params_parse(argc, argv, params, options)) {
         return 1;
     }
 

examples/llava/minicpmv-cli.cpp

@@ -1,9 +1,11 @@
-#include "ggml.h"
+#include "arg.h"
 #include "log.h"
 #include "common.h"
+#include "sampling.h"
 #include "clip.h"
 #include "llava.h"
 #include "llama.h"
+#include "ggml.h"
 
 #include <cstdio>
 #include <cstdlib>
@@ -253,8 +255,7 @@ int main(int argc, char ** argv) {
 
     gpt_params params;
 
-    auto options = gpt_params_parser_init(params, LLAMA_EXAMPLE_COMMON, show_additional_info);
+    if (!gpt_params_parse(argc, argv, params, LLAMA_EXAMPLE_COMMON, show_additional_info)) {
-    if (!gpt_params_parse(argc, argv, params, options)) {
         return 1;
     }
 

examples/lookahead/lookahead.cpp

@@ -1,4 +1,6 @@
+#include "arg.h"
 #include "common.h"
+#include "sampling.h"
 #include "llama.h"
 
 #include <cstdio>
@@ -36,8 +38,7 @@ struct ngram_container {
 int main(int argc, char ** argv) {
     gpt_params params;
 
-    auto options = gpt_params_parser_init(params, LLAMA_EXAMPLE_COMMON);
+    if (!gpt_params_parse(argc, argv, params, LLAMA_EXAMPLE_COMMON)) {
-    if (!gpt_params_parse(argc, argv, params, options)) {
         return 1;
     }
 

examples/lookup/lookup-create.cpp

@@ -1,7 +1,8 @@
-#include "ggml.h"
+#include "arg.h"
-#include "llama.h"
 #include "common.h"
 #include "ngram-cache.h"
+#include "ggml.h"
+#include "llama.h"
 
 #include <cstdint>
 #include <fstream>
@@ -13,8 +14,7 @@
 int main(int argc, char ** argv){
     gpt_params params;
 
-    auto options = gpt_params_parser_init(params, LLAMA_EXAMPLE_COMMON);
+    if (!gpt_params_parse(argc, argv, params, LLAMA_EXAMPLE_LOOKUP)) {
-    if (!gpt_params_parse(argc, argv, params, options)) {
         return 1;
     }
 
@@ -40,4 +40,6 @@ int main(int argc, char ** argv){
     fprintf(stderr, "%s: hashing done, writing file to %s\n", __func__, params.lookup_cache_static.c_str());
 
     llama_ngram_cache_save(ngram_cache, params.lookup_cache_static);
+
+    return 0;
 }

examples/lookup/lookup-stats.cpp

@@ -1,8 +1,9 @@
-#include "ggml.h"
+#include "arg.h"
 #include "common.h"
-#include "llama.h"
 #include "log.h"
 #include "ngram-cache.h"
+#include "llama.h"
+#include "ggml.h"
 
 #include <cmath>
 #include <cstdint>
@@ -15,8 +16,7 @@
 int main(int argc, char ** argv){
     gpt_params params;
 
-    auto options = gpt_params_parser_init(params, LLAMA_EXAMPLE_COMMON);
+    if (!gpt_params_parse(argc, argv, params, LLAMA_EXAMPLE_LOOKUP)) {
-    if (!gpt_params_parse(argc, argv, params, options)) {
         return 1;
     }
 

examples/lookup/lookup.cpp

@@ -1,7 +1,9 @@
+#include "arg.h"
 #include "ggml.h"
-#include "llama.h"
 #include "common.h"
 #include "ngram-cache.h"
+#include "sampling.h"
+#include "llama.h"
 
 #include <cstdint>
 #include <cstdio>
@@ -12,8 +14,7 @@
 int main(int argc, char ** argv){
     gpt_params params;
 
-    auto options = gpt_params_parser_init(params, LLAMA_EXAMPLE_COMMON);
+    if (!gpt_params_parse(argc, argv, params, LLAMA_EXAMPLE_LOOKUP)) {
-    if (!gpt_params_parse(argc, argv, params, options)) {
         return 1;
     }
 

examples/main/main.cpp

@@ -1,6 +1,7 @@
+#include "arg.h"
 #include "common.h"
-
 #include "console.h"
+#include "sampling.h"
 #include "llama.h"
 
 #include <cassert>
@@ -138,9 +139,7 @@ static std::string chat_add_and_format(struct llama_model * model, std::vector<l
 int main(int argc, char ** argv) {
     gpt_params params;
     g_params = &params;
-    auto options = gpt_params_parser_init(params, LLAMA_EXAMPLE_MAIN, print_usage);
+    if (!gpt_params_parse(argc, argv, params, LLAMA_EXAMPLE_MAIN, print_usage)) {
-
-    if (!gpt_params_parse(argc, argv, params, options)) {
         return 1;
     }
 

examples/parallel/parallel.cpp

@@ -1,7 +1,9 @@
 // A basic application simulating a server with multiple clients.
 // The clients submit requests to the server and they are processed in parallel.
 
+#include "arg.h"
 #include "common.h"
+#include "sampling.h"
 #include "llama.h"
 
 #include <cmath>
@@ -100,8 +102,7 @@ int main(int argc, char ** argv) {
 
     gpt_params params;
 
-    auto options = gpt_params_parser_init(params, LLAMA_EXAMPLE_COMMON);
+    if (!gpt_params_parse(argc, argv, params, LLAMA_EXAMPLE_PARALLEL)) {
-    if (!gpt_params_parse(argc, argv, params, options)) {
         return 1;
     }
 

examples/passkey/passkey.cpp

@@ -1,3 +1,4 @@
+#include "arg.h"
 #include "common.h"
 #include "llama.h"
 
@@ -19,8 +20,7 @@ int main(int argc, char ** argv) {
     params.n_keep = 32;
     params.i_pos  = -1;
 
-    auto options = gpt_params_parser_init(params, LLAMA_EXAMPLE_PASSKEY, print_usage);
+    if (!gpt_params_parse(argc, argv, params, LLAMA_EXAMPLE_PASSKEY, print_usage)) {
-    if (!gpt_params_parse(argc, argv, params, options)) {
         return 1;
     }
 

examples/perplexity/perplexity.cpp

@@ -1,18 +1,19 @@
+#include "arg.h"
 #include "common.h"
 #include "llama.h"
 
+#include <array>
+#include <atomic>
 #include <cmath>
 #include <cstdio>
 #include <cstring>
 #include <ctime>
+#include <fstream>
+#include <mutex>
+#include <random>
 #include <sstream>
 #include <thread>
-#include <mutex>
-#include <atomic>
 #include <vector>
-#include <array>
-#include <fstream>
-#include <sstream>
 
 #if defined(_MSC_VER)
 #pragma warning(disable: 4244 4267) // possible loss of data
@@ -1967,8 +1968,7 @@ int main(int argc, char ** argv) {
     params.n_ctx = 512;
     params.logits_all = true;
 
-    auto options = gpt_params_parser_init(params, LLAMA_EXAMPLE_PERPLEXITY);
+    if (!gpt_params_parse(argc, argv, params, LLAMA_EXAMPLE_PERPLEXITY)) {
-    if (!gpt_params_parse(argc, argv, params, options)) {
         return 1;
     }
 

examples/retrieval/retrieval.cpp

@@ -1,3 +1,4 @@
+#include "arg.h"
 #include "common.h"
 #include "llama.h"
 
@@ -111,8 +112,7 @@ static void batch_decode(llama_context * ctx, llama_batch & batch, float * outpu
 int main(int argc, char ** argv) {
     gpt_params params;
 
-    auto options = gpt_params_parser_init(params, LLAMA_EXAMPLE_RETRIEVAL, print_usage);
+    if (!gpt_params_parse(argc, argv, params, LLAMA_EXAMPLE_RETRIEVAL, print_usage)) {
-    if (!gpt_params_parse(argc, argv, params, options)) {
         return 1;
     }
 

examples/save-load-state/save-load-state.cpp

@@ -1,3 +1,4 @@
+#include "arg.h"
 #include "common.h"
 #include "llama.h"
 
@@ -10,8 +11,7 @@ int main(int argc, char ** argv) {
     params.prompt = "The quick brown fox";
     params.sparams.seed = 1234;
 
-    auto options = gpt_params_parser_init(params, LLAMA_EXAMPLE_COMMON);
+    if (!gpt_params_parse(argc, argv, params, LLAMA_EXAMPLE_COMMON)) {
-    if (!gpt_params_parse(argc, argv, params, options)) {
         return 1;
     }
 

examples/server/README.md

@@ -23,36 +23,32 @@ The project is under active development, and we are [looking for feedback and co
 | `--version` | show version and build info |
 | `-v, --verbose` | print verbose information |
 | `--verbosity N` | set specific verbosity level (default: 0) |
-| `--verbose-prompt` | print a verbose prompt before generation (default: false) |
-| `--no-display-prompt` | don't print prompt at generation (default: false) |
-| `-s, --seed SEED` | RNG seed (default: -1, use random seed for < 0) |
 | `-t, --threads N` | number of threads to use during generation (default: -1)<br/>(env: LLAMA_ARG_THREADS) |
 | `-tb, --threads-batch N` | number of threads to use during batch and prompt processing (default: same as --threads) |
 | `-C, --cpu-mask M` | CPU affinity mask: arbitrarily long hex. Complements cpu-range (default: "") |
 | `-Cr, --cpu-range lo-hi` | range of CPUs for affinity. Complements --cpu-mask |
 | `--cpu-strict <0\|1>` | use strict CPU placement (default: 0)<br/> |
+| `--prio N` | set process/thread priority : 0-normal, 1-medium, 2-high, 3-realtime (default: 0)<br/> |
 | `--poll <0...100>` | use polling level to wait for work (0 - no polling, default: 50)<br/> |
 | `-Cb, --cpu-mask-batch M` | CPU affinity mask: arbitrarily long hex. Complements cpu-range-batch (default: same as --cpu-mask) |
 | `-Crb, --cpu-range-batch lo-hi` | ranges of CPUs for affinity. Complements --cpu-mask-batch |
 | `--cpu-strict-batch <0\|1>` | use strict CPU placement (default: same as --cpu-strict) |
+| `--prio-batch N` | set process/thread priority : 0-normal, 1-medium, 2-high, 3-realtime (default: 0)<br/> |
 | `--poll-batch <0\|1>` | use polling to wait for work (default: same as --poll) |
-| `-lcs, --lookup-cache-static FNAME` | path to static lookup cache to use for lookup decoding (not updated by generation) |
-| `-lcd, --lookup-cache-dynamic FNAME` | path to dynamic lookup cache to use for lookup decoding (updated by generation) |
 | `-c, --ctx-size N` | size of the prompt context (default: 0, 0 = loaded from model)<br/>(env: LLAMA_ARG_CTX_SIZE) |
 | `-n, --predict, --n-predict N` | number of tokens to predict (default: -1, -1 = infinity, -2 = until context filled)<br/>(env: LLAMA_ARG_N_PREDICT) |
 | `-b, --batch-size N` | logical maximum batch size (default: 2048)<br/>(env: LLAMA_ARG_BATCH) |
 | `-ub, --ubatch-size N` | physical maximum batch size (default: 512)<br/>(env: LLAMA_ARG_UBATCH) |
 | `--keep N` | number of tokens to keep from the initial prompt (default: 0, -1 = all) |
-| `--chunks N` | max number of chunks to process (default: -1, -1 = all) |
 | `-fa, --flash-attn` | enable Flash Attention (default: disabled)<br/>(env: LLAMA_ARG_FLASH_ATTN) |
 | `-p, --prompt PROMPT` | prompt to start generation with |
 | `-f, --file FNAME` | a file containing the prompt (default: none) |
-| `--in-file FNAME` | an input file (repeat to specify multiple files) |
 | `-bf, --binary-file FNAME` | binary file containing the prompt (default: none) |
 | `-e, --escape` | process escapes sequences (\n, \r, \t, \', \", \\) (default: true) |
 | `--no-escape` | do not process escape sequences |
 | `--spm-infill` | use Suffix/Prefix/Middle pattern for infill (instead of Prefix/Suffix/Middle) as some models prefer this. (default: disabled) |
-| `--samplers SAMPLERS` | samplers that will be used for generation in the order, separated by ';'<br/>(default: top_k;tfs_z;typical_p;top_p;min_p;temperature) |
+| `--samplers SAMPLERS` | samplers that will be used for generation in the order, separated by ';'<br/>(default: top_k;tfs_z;typ_p;top_p;min_p;temperature) |
+| `-s, --seed SEED` | RNG seed (default: -1, use random seed for < 0) |
 | `--sampling-seq SEQUENCE` | simplified sequence for samplers that will be used (default: kfypmt) |
 | `--ignore-eos` | ignore end of stream token and continue generating (implies --logit-bias EOS-inf) |
 | `--penalize-nl` | penalize newline tokens (default: false) |
@@ -92,13 +88,12 @@ The project is under active development, and we are [looking for feedback and co
 | `-ctv, --cache-type-v TYPE` | KV cache data type for V (default: f16) |
 | `-dt, --defrag-thold N` | KV cache defragmentation threshold (default: -1.0, < 0 - disabled)<br/>(env: LLAMA_ARG_DEFRAG_THOLD) |
 | `-np, --parallel N` | number of parallel sequences to decode (default: 1) |
-| `-ns, --sequences N` | number of sequences to decode (default: 1) |
 | `-cb, --cont-batching` | enable continuous batching (a.k.a dynamic batching) (default: enabled)<br/>(env: LLAMA_ARG_CONT_BATCHING) |
 | `-nocb, --no-cont-batching` | disable continuous batching<br/>(env: LLAMA_ARG_NO_CONT_BATCHING) |
 | `--mlock` | force system to keep model in RAM rather than swapping or compressing |
 | `--no-mmap` | do not memory-map model (slower load but may reduce pageouts if not using mlock) |
 | `--numa TYPE` | attempt optimizations that help on some NUMA systems<br/>- distribute: spread execution evenly over all nodes<br/>- isolate: only spawn threads on CPUs on the node that execution started on<br/>- numactl: use the CPU map provided by numactl<br/>if run without this previously, it is recommended to drop the system page cache before using this<br/>see https://github.com/ggerganov/llama.cpp/issues/1437 |
-| `-ngl, --gpu-layers N` | number of layers to store in VRAM<br/>(env: LLAMA_ARG_N_GPU_LAYERS) |
+| `-ngl, --gpu-layers, --n-gpu-layers N` | number of layers to store in VRAM<br/>(env: LLAMA_ARG_N_GPU_LAYERS) |
 | `-sm, --split-mode {none,layer,row}` | how to split the model across multiple GPUs, one of:<br/>- none: use one GPU only<br/>- layer (default): split layers and KV across GPUs<br/>- row: split rows across GPUs |
 | `-ts, --tensor-split N0,N1,N2,...` | fraction of the model to offload to each GPU, comma-separated list of proportions, e.g. 3,1 |
 | `-mg, --main-gpu INDEX` | the GPU to use for the model (with split-mode = none), or for intermediate results and KV (with split-mode = row) (default: 0) |
@@ -109,7 +104,7 @@ The project is under active development, and we are [looking for feedback and co
 | `--control-vector FNAME` | add a control vector<br/>note: this argument can be repeated to add multiple control vectors |
 | `--control-vector-scaled FNAME SCALE` | add a control vector with user defined scaling SCALE<br/>note: this argument can be repeated to add multiple scaled control vectors |
 | `--control-vector-layer-range START END` | layer range to apply the control vector(s) to, start and end inclusive |
-| `-a, --alias STRING` | set alias for model name (to be used by REST API)<br/>(env: LLAMA_ARG_MODEL) |
+| `-a, --alias STRING` | set alias for model name (to be used by REST API) |
 | `-m, --model FNAME` | model path (default: `models/$filename` with filename from `--hf-file` or `--model-url` if set, otherwise models/7B/ggml-model-f16.gguf)<br/>(env: LLAMA_ARG_MODEL) |
 | `-mu, --model-url MODEL_URL` | model download url (default: unused)<br/>(env: LLAMA_ARG_MODEL_URL) |
 | `-hfr, --hf-repo REPO` | Hugging Face model repository (default: unused)<br/>(env: LLAMA_ARG_HF_REPO) |
@@ -123,7 +118,7 @@ The project is under active development, and we are [looking for feedback and co
 | `--api-key-file FNAME` | path to file containing API keys (default: none) |
 | `--ssl-key-file FNAME` | path to file a PEM-encoded SSL private key |
 | `--ssl-cert-file FNAME` | path to file a PEM-encoded SSL certificate |
-| `--timeout N` | server read/write timeout in seconds (default: 600) |
+| `-to, --timeout N` | server read/write timeout in seconds (default: 600) |
 | `--threads-http N` | number of threads used to process HTTP requests (default: -1)<br/>(env: LLAMA_ARG_THREADS_HTTP) |
 | `-spf, --system-prompt-file FNAME` | set a file to load a system prompt (initial prompt of all slots), this is useful for chat applications |
 | `--log-format {text, json}` | log output format: json or text (default: json) |

examples/server/server.cpp

@@ -1,6 +1,8 @@
 #include "utils.hpp"
 
+#include "arg.h"
 #include "common.h"
+#include "sampling.h"
 #include "json-schema-to-grammar.h"
 #include "llama.h"
 
@@ -2423,8 +2425,7 @@ int main(int argc, char ** argv) {
     // own arguments required by this example
     gpt_params params;
 
-    auto options = gpt_params_parser_init(params, LLAMA_EXAMPLE_SERVER);
+    if (!gpt_params_parse(argc, argv, params, LLAMA_EXAMPLE_SERVER)) {
-    if (!gpt_params_parse(argc, argv, params, options)) {
         return 1;
     }
 

examples/simple/simple.cpp

@@ -1,3 +1,4 @@
+#include "arg.h"
 #include "common.h"
 #include "llama.h"
 
@@ -18,8 +19,7 @@ int main(int argc, char ** argv) {
     params.prompt = "Hello my name is";
     params.n_predict = 32;
 
-    auto options = gpt_params_parser_init(params, LLAMA_EXAMPLE_COMMON, print_usage);
+    if (!gpt_params_parse(argc, argv, params, LLAMA_EXAMPLE_COMMON, print_usage)) {
-    if (!gpt_params_parse(argc, argv, params, options)) {
         return 1;
     }
 

examples/speculative/speculative.cpp

@@ -1,11 +1,13 @@
+#include "arg.h"
 #include "common.h"
+#include "sampling.h"
 #include "llama.h"
 
-#include <cmath>
 #include <cstdio>
 #include <string>
 #include <vector>
 #include <set>
+#include <random>
 
 #define SPEC_VOCAB_MAX_SIZE_DIFFERENCE  100
 #define SPEC_VOCAB_CHECK_START_TOKEN_ID 5
@@ -27,8 +29,7 @@ struct seq_draft {
 int main(int argc, char ** argv) {
     gpt_params params;
 
-    auto options = gpt_params_parser_init(params, LLAMA_EXAMPLE_SPECULATIVE);
+    if (!gpt_params_parse(argc, argv, params, LLAMA_EXAMPLE_SPECULATIVE)) {
-    if (!gpt_params_parse(argc, argv, params, options)) {
         return 1;
     }
 

tests/test-arg-parser.cpp

@@ -1,3 +1,6 @@
+#include "arg.h"
+#include "common.h"
+
 #include <string>
 #include <vector>
 #include <sstream>
@@ -6,18 +9,16 @@
 #undef NDEBUG
 #include <cassert>
 
-#include "common.h"
-
 int main(void) {
     gpt_params params;
 
     printf("test-arg-parser: make sure there is no duplicated arguments in any examples\n\n");
     for (int ex = 0; ex < LLAMA_EXAMPLE_COUNT; ex++) {
         try {
-            auto options = gpt_params_parser_init(params, (enum llama_example)ex);
+            auto ctx_arg = gpt_params_parser_init(params, (enum llama_example)ex);
             std::unordered_set<std::string> seen_args;
             std::unordered_set<std::string> seen_env_vars;
-            for (const auto & opt : options) {
+            for (const auto & opt : ctx_arg.options) {
                 // check for args duplications
                 for (const auto & arg : opt.args) {
                     if (seen_args.find(arg) == seen_args.end()) {
@@ -52,40 +53,51 @@ int main(void) {
     };
 
     std::vector<std::string> argv;
-    auto options = gpt_params_parser_init(params, LLAMA_EXAMPLE_COMMON);
 
     printf("test-arg-parser: test invalid usage\n\n");
 
+    // missing value
     argv = {"binary_name", "-m"};
-    assert(false == gpt_params_parse(argv.size(), list_str_to_char(argv).data(), params, options));
+    assert(false == gpt_params_parse(argv.size(), list_str_to_char(argv).data(), params, LLAMA_EXAMPLE_COMMON));
 
+    // wrong value (int)
     argv = {"binary_name", "-ngl", "hello"};
-    assert(false == gpt_params_parse(argv.size(), list_str_to_char(argv).data(), params, options));
+    assert(false == gpt_params_parse(argv.size(), list_str_to_char(argv).data(), params, LLAMA_EXAMPLE_COMMON));
 
+    // wrong value (enum)
     argv = {"binary_name", "-sm", "hello"};
-    assert(false == gpt_params_parse(argv.size(), list_str_to_char(argv).data(), params, options));
+    assert(false == gpt_params_parse(argv.size(), list_str_to_char(argv).data(), params, LLAMA_EXAMPLE_COMMON));
+
+    // non-existence arg in specific example (--draft cannot be used outside llama-speculative)
+    argv = {"binary_name", "--draft", "123"};
+    assert(false == gpt_params_parse(argv.size(), list_str_to_char(argv).data(), params, LLAMA_EXAMPLE_SERVER));
 
 
     printf("test-arg-parser: test valid usage\n\n");
 
     argv = {"binary_name", "-m", "model_file.gguf"};
-    assert(true == gpt_params_parse(argv.size(), list_str_to_char(argv).data(), params, options));
+    assert(true == gpt_params_parse(argv.size(), list_str_to_char(argv).data(), params, LLAMA_EXAMPLE_COMMON));
     assert(params.model == "model_file.gguf");
 
     argv = {"binary_name", "-t", "1234"};
-    assert(true == gpt_params_parse(argv.size(), list_str_to_char(argv).data(), params, options));
+    assert(true == gpt_params_parse(argv.size(), list_str_to_char(argv).data(), params, LLAMA_EXAMPLE_COMMON));
     assert(params.cpuparams.n_threads == 1234);
 
     argv = {"binary_name", "--verbose"};
-    assert(true == gpt_params_parse(argv.size(), list_str_to_char(argv).data(), params, options));
+    assert(true == gpt_params_parse(argv.size(), list_str_to_char(argv).data(), params, LLAMA_EXAMPLE_COMMON));
     assert(params.verbosity == 1);
 
     argv = {"binary_name", "-m", "abc.gguf", "--predict", "6789", "--batch-size", "9090"};
-    assert(true == gpt_params_parse(argv.size(), list_str_to_char(argv).data(), params, options));
+    assert(true == gpt_params_parse(argv.size(), list_str_to_char(argv).data(), params, LLAMA_EXAMPLE_COMMON));
     assert(params.model == "abc.gguf");
     assert(params.n_predict == 6789);
     assert(params.n_batch == 9090);
 
+    // --draft cannot be used outside llama-speculative
+    argv = {"binary_name", "--draft", "123"};
+    assert(true == gpt_params_parse(argv.size(), list_str_to_char(argv).data(), params, LLAMA_EXAMPLE_SPECULATIVE));
+    assert(params.n_draft == 123);
+
 // skip this part on windows, because setenv is not supported
 #ifdef _WIN32
     printf("test-arg-parser: skip on windows build\n");
@@ -94,12 +106,12 @@ int main(void) {
 
     setenv("LLAMA_ARG_THREADS", "blah", true);
     argv = {"binary_name"};
-    assert(false == gpt_params_parse(argv.size(), list_str_to_char(argv).data(), params, options));
+    assert(false == gpt_params_parse(argv.size(), list_str_to_char(argv).data(), params, LLAMA_EXAMPLE_COMMON));
 
     setenv("LLAMA_ARG_MODEL", "blah.gguf", true);
     setenv("LLAMA_ARG_THREADS", "1010", true);
     argv = {"binary_name"};
-    assert(true == gpt_params_parse(argv.size(), list_str_to_char(argv).data(), params, options));
+    assert(true == gpt_params_parse(argv.size(), list_str_to_char(argv).data(), params, LLAMA_EXAMPLE_COMMON));
     assert(params.model == "blah.gguf");
     assert(params.cpuparams.n_threads == 1010);
 
@@ -109,7 +121,7 @@ int main(void) {
     setenv("LLAMA_ARG_MODEL", "blah.gguf", true);
     setenv("LLAMA_ARG_THREADS", "1010", true);
     argv = {"binary_name", "-m", "overwritten.gguf"};
-    assert(true == gpt_params_parse(argv.size(), list_str_to_char(argv).data(), params, options));
+    assert(true == gpt_params_parse(argv.size(), list_str_to_char(argv).data(), params, LLAMA_EXAMPLE_COMMON));
     assert(params.model == "overwritten.gguf");
     assert(params.cpuparams.n_threads == 1010);
 #endif // _WIN32