npm - @fugood/llama.node - Versions diffs - 0.3.2 → 0.3.3 - Mend

@fugood/llama.node 0.3.2 → 0.3.3

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Files changed (190) hide show

package/CMakeLists.txt +2 -0
package/bin/darwin/arm64/llama-node.node +0 -0
package/bin/darwin/x64/llama-node.node +0 -0
package/bin/linux/arm64/llama-node.node +0 -0
package/bin/linux/x64/llama-node.node +0 -0
package/bin/linux-vulkan/arm64/llama-node.node +0 -0
package/bin/linux-vulkan/x64/llama-node.node +0 -0
package/bin/win32/arm64/llama-node.node +0 -0
package/bin/win32/arm64/node.lib +0 -0
package/bin/win32/x64/llama-node.node +0 -0
package/bin/win32/x64/node.lib +0 -0
package/bin/win32-vulkan/arm64/llama-node.node +0 -0
package/bin/win32-vulkan/arm64/node.lib +0 -0
package/bin/win32-vulkan/x64/llama-node.node +0 -0
package/bin/win32-vulkan/x64/node.lib +0 -0
package/package.json +1 -1
package/src/DetokenizeWorker.cpp +1 -1
package/src/EmbeddingWorker.cpp +2 -2
package/src/LlamaCompletionWorker.cpp +8 -8
package/src/LlamaCompletionWorker.h +2 -2
package/src/LlamaContext.cpp +8 -9
package/src/TokenizeWorker.cpp +1 -1
package/src/common.hpp +4 -4
package/src/llama.cpp/.github/workflows/build.yml +43 -9
package/src/llama.cpp/.github/workflows/docker.yml +3 -0
package/src/llama.cpp/CMakeLists.txt +7 -4
package/src/llama.cpp/cmake/arm64-apple-clang.cmake +16 -0
package/src/llama.cpp/common/CMakeLists.txt +0 -2
package/src/llama.cpp/common/arg.cpp +642 -607
package/src/llama.cpp/common/arg.h +22 -22
package/src/llama.cpp/common/common.cpp +79 -281
package/src/llama.cpp/common/common.h +130 -100
package/src/llama.cpp/common/json-schema-to-grammar.cpp +1 -1
package/src/llama.cpp/common/log.cpp +50 -50
package/src/llama.cpp/common/log.h +18 -18
package/src/llama.cpp/common/ngram-cache.cpp +36 -36
package/src/llama.cpp/common/ngram-cache.h +19 -19
package/src/llama.cpp/common/sampling.cpp +116 -108
package/src/llama.cpp/common/sampling.h +20 -20
package/src/llama.cpp/docs/build.md +37 -17
package/src/llama.cpp/examples/CMakeLists.txt +1 -1
package/src/llama.cpp/examples/batched/batched.cpp +14 -14
package/src/llama.cpp/examples/batched-bench/batched-bench.cpp +10 -11
package/src/llama.cpp/examples/convert-llama2c-to-ggml/convert-llama2c-to-ggml.cpp +1 -1
package/src/llama.cpp/examples/cvector-generator/cvector-generator.cpp +9 -9
package/src/llama.cpp/examples/embedding/embedding.cpp +12 -12
package/src/llama.cpp/examples/eval-callback/eval-callback.cpp +8 -8
package/src/llama.cpp/examples/export-lora/export-lora.cpp +5 -5
package/src/llama.cpp/examples/gen-docs/gen-docs.cpp +7 -7
package/src/llama.cpp/examples/gritlm/gritlm.cpp +18 -18
package/src/llama.cpp/examples/imatrix/imatrix.cpp +20 -11
package/src/llama.cpp/examples/infill/infill.cpp +40 -86
package/src/llama.cpp/examples/llama-bench/llama-bench.cpp +42 -151
package/src/llama.cpp/examples/llama.android/llama/build.gradle.kts +1 -0
package/src/llama.cpp/examples/llama.android/llama/src/main/cpp/llama-android.cpp +11 -14
package/src/llama.cpp/examples/llava/clip.cpp +1 -0
package/src/llama.cpp/examples/llava/llava-cli.cpp +23 -23
package/src/llama.cpp/examples/llava/llava.cpp +37 -3
package/src/llama.cpp/examples/llava/minicpmv-cli.cpp +21 -21
package/src/llama.cpp/examples/lookahead/lookahead.cpp +26 -26
package/src/llama.cpp/examples/lookup/lookup-create.cpp +7 -7
package/src/llama.cpp/examples/lookup/lookup-merge.cpp +4 -4
package/src/llama.cpp/examples/lookup/lookup-stats.cpp +14 -14
package/src/llama.cpp/examples/lookup/lookup.cpp +29 -29
package/src/llama.cpp/examples/main/main.cpp +64 -109
package/src/llama.cpp/examples/parallel/parallel.cpp +18 -19
package/src/llama.cpp/examples/passkey/passkey.cpp +14 -14
package/src/llama.cpp/examples/perplexity/perplexity.cpp +99 -120
package/src/llama.cpp/examples/quantize-stats/quantize-stats.cpp +10 -9
package/src/llama.cpp/examples/retrieval/retrieval.cpp +13 -13
package/src/llama.cpp/examples/rpc/rpc-server.cpp +3 -1
package/src/llama.cpp/examples/save-load-state/save-load-state.cpp +34 -17
package/src/llama.cpp/examples/server/CMakeLists.txt +4 -13
package/src/llama.cpp/examples/server/server.cpp +553 -691
package/src/llama.cpp/examples/server/utils.hpp +312 -25
package/src/llama.cpp/examples/simple/CMakeLists.txt +1 -1
package/src/llama.cpp/examples/simple/simple.cpp +128 -96
package/src/llama.cpp/examples/simple-chat/CMakeLists.txt +5 -0
package/src/llama.cpp/examples/simple-chat/simple-chat.cpp +197 -0
package/src/llama.cpp/examples/speculative/speculative.cpp +54 -51
package/src/llama.cpp/examples/tokenize/tokenize.cpp +2 -2
package/src/llama.cpp/ggml/CMakeLists.txt +15 -9
package/src/llama.cpp/ggml/include/ggml-amx.h +25 -0
package/src/llama.cpp/ggml/include/ggml-backend.h +46 -33
package/src/llama.cpp/ggml/include/ggml-blas.h +5 -3
package/src/llama.cpp/ggml/include/ggml-cann.h +9 -7
package/src/llama.cpp/ggml/include/ggml-cpp.h +38 -0
package/src/llama.cpp/ggml/include/ggml-cpu.h +177 -0
package/src/llama.cpp/ggml/include/ggml-cuda.h +12 -12
package/src/llama.cpp/ggml/include/ggml-kompute.h +7 -3
package/src/llama.cpp/ggml/include/ggml-metal.h +11 -7
package/src/llama.cpp/ggml/include/ggml-opt.h +216 -0
package/src/llama.cpp/ggml/include/ggml-rpc.h +9 -5
package/src/llama.cpp/ggml/include/ggml-sycl.h +18 -11
package/src/llama.cpp/ggml/include/ggml-vulkan.h +10 -8
package/src/llama.cpp/ggml/include/ggml.h +53 -393
package/src/llama.cpp/ggml/src/CMakeLists.txt +66 -1149
package/src/llama.cpp/ggml/src/ggml-aarch64.c +46 -3126
package/src/llama.cpp/ggml/src/ggml-aarch64.h +0 -20
package/src/llama.cpp/ggml/src/ggml-alloc.c +23 -27
package/src/llama.cpp/ggml/src/ggml-amx/CMakeLists.txt +107 -0
package/src/llama.cpp/ggml/src/ggml-amx/common.h +94 -0
package/src/llama.cpp/ggml/src/ggml-amx/ggml-amx.cpp +446 -0
package/src/llama.cpp/ggml/src/ggml-amx/mmq.cpp +2510 -0
package/src/llama.cpp/ggml/src/ggml-amx/mmq.h +17 -0
package/src/llama.cpp/ggml/src/ggml-backend-impl.h +6 -25
package/src/llama.cpp/ggml/src/ggml-backend-reg.cpp +195 -0
package/src/llama.cpp/ggml/src/ggml-backend.cpp +303 -864
package/src/llama.cpp/ggml/src/ggml-blas/CMakeLists.txt +91 -0
package/src/llama.cpp/ggml/src/{ggml-blas.cpp → ggml-blas/ggml-blas.cpp} +213 -65
package/src/llama.cpp/ggml/src/ggml-cann/CMakeLists.txt +46 -0
package/src/llama.cpp/ggml/src/{ggml-cann.cpp → ggml-cann/ggml-cann.cpp} +255 -149
package/src/llama.cpp/ggml/src/ggml-cpu/CMakeLists.txt +261 -0
package/src/llama.cpp/ggml/src/ggml-cpu/ggml-cpu-aarch64.c +3560 -0
package/src/llama.cpp/ggml/src/ggml-cpu/ggml-cpu-aarch64.h +30 -0
package/src/llama.cpp/ggml/src/{ggml-cpu-impl.h → ggml-cpu/ggml-cpu-impl.h} +0 -243
package/src/llama.cpp/ggml/src/ggml-cpu/ggml-cpu-quants.c +10822 -0
package/src/llama.cpp/ggml/src/ggml-cpu/ggml-cpu-quants.h +63 -0
package/src/llama.cpp/ggml/src/ggml-cpu/ggml-cpu.c +13970 -0
package/src/llama.cpp/ggml/src/ggml-cpu/ggml-cpu.cpp +663 -0
package/src/llama.cpp/ggml/src/{llamafile → ggml-cpu/llamafile}/sgemm.cpp +667 -1
package/src/llama.cpp/ggml/src/ggml-cuda/CMakeLists.txt +155 -0
package/src/llama.cpp/ggml/src/ggml-hip/CMakeLists.txt +106 -0
package/src/llama.cpp/ggml/src/ggml-impl.h +366 -16
package/src/llama.cpp/ggml/src/ggml-kompute/CMakeLists.txt +162 -0
package/src/llama.cpp/ggml/src/{ggml-kompute.cpp → ggml-kompute/ggml-kompute.cpp} +238 -72
package/src/llama.cpp/ggml/src/ggml-metal/CMakeLists.txt +108 -0
package/src/llama.cpp/ggml/src/ggml-metal/ggml-metal-impl.h +249 -0
package/src/llama.cpp/ggml/src/ggml-musa/CMakeLists.txt +100 -0
package/src/llama.cpp/ggml/src/ggml-opt.cpp +867 -0
package/src/llama.cpp/ggml/src/ggml-quants.c +187 -10692
package/src/llama.cpp/ggml/src/ggml-quants.h +78 -125
package/src/llama.cpp/ggml/src/ggml-rpc/CMakeLists.txt +11 -0
package/src/llama.cpp/ggml/src/{ggml-rpc.cpp → ggml-rpc/ggml-rpc.cpp} +475 -300
package/src/llama.cpp/ggml/src/ggml-sycl/CMakeLists.txt +81 -0
package/src/llama.cpp/ggml/src/ggml-sycl/backend.hpp +3 -0
package/src/llama.cpp/ggml/src/ggml-sycl/common.cpp +40 -0
package/src/llama.cpp/ggml/src/ggml-sycl/common.hpp +258 -0
package/src/llama.cpp/ggml/src/ggml-sycl/concat.cpp +1 -0
package/src/llama.cpp/ggml/src/ggml-sycl/dpct/helper.hpp +2 -22
package/src/llama.cpp/ggml/src/ggml-sycl/element_wise.cpp +1011 -0
package/src/llama.cpp/ggml/src/ggml-sycl/element_wise.hpp +76 -0
package/src/llama.cpp/ggml/src/{ggml-sycl.cpp → ggml-sycl/ggml-sycl.cpp} +3584 -4142
package/src/llama.cpp/ggml/src/ggml-sycl/mmvq.cpp +69 -67
package/src/llama.cpp/ggml/src/ggml-sycl/norm.cpp +3 -3
package/src/llama.cpp/ggml/src/ggml-sycl/outprod.cpp +56 -0
package/src/llama.cpp/ggml/src/ggml-sycl/outprod.hpp +11 -0
package/src/llama.cpp/ggml/src/ggml-sycl/presets.hpp +6 -0
package/src/llama.cpp/ggml/src/ggml-sycl/vecdotq.hpp +4 -4
package/src/llama.cpp/ggml/src/ggml-sycl/wkv6.cpp +138 -0
package/src/llama.cpp/ggml/src/ggml-sycl/wkv6.hpp +10 -0
package/src/llama.cpp/ggml/src/ggml-threading.cpp +12 -0
package/src/llama.cpp/ggml/src/ggml-threading.h +12 -0
package/src/llama.cpp/ggml/src/ggml-vulkan/CMakeLists.txt +78 -0
package/src/llama.cpp/ggml/src/{ggml-vulkan.cpp → ggml-vulkan/ggml-vulkan.cpp} +555 -623
package/src/llama.cpp/ggml/src/{vulkan-shaders → ggml-vulkan/vulkan-shaders}/vulkan-shaders-gen.cpp +125 -206
package/src/llama.cpp/ggml/src/ggml.c +4032 -19890
package/src/llama.cpp/include/llama.h +67 -33
package/src/llama.cpp/pocs/vdot/q8dot.cpp +4 -3
package/src/llama.cpp/pocs/vdot/vdot.cpp +8 -7
package/src/llama.cpp/src/CMakeLists.txt +2 -1
package/src/llama.cpp/src/llama-sampling.cpp +745 -105
package/src/llama.cpp/src/llama-sampling.h +21 -2
package/src/llama.cpp/src/llama-vocab.cpp +49 -9
package/src/llama.cpp/src/llama-vocab.h +35 -11
package/src/llama.cpp/src/llama.cpp +2636 -2406
package/src/llama.cpp/src/unicode-data.cpp +2 -2
package/src/llama.cpp/tests/CMakeLists.txt +1 -2
package/src/llama.cpp/tests/test-arg-parser.cpp +14 -14
package/src/llama.cpp/tests/test-backend-ops.cpp +185 -60
package/src/llama.cpp/tests/test-barrier.cpp +1 -0
package/src/llama.cpp/tests/test-chat-template.cpp +9 -5
package/src/llama.cpp/tests/test-json-schema-to-grammar.cpp +17 -4
package/src/llama.cpp/tests/test-log.cpp +2 -2
package/src/llama.cpp/tests/test-opt.cpp +853 -142
package/src/llama.cpp/tests/test-quantize-fns.cpp +22 -19
package/src/llama.cpp/tests/test-quantize-perf.cpp +16 -14
package/src/llama.cpp/tests/test-rope.cpp +1 -0
package/src/llama.cpp/tests/test-sampling.cpp +162 -137
package/src/llama.cpp/tests/test-tokenizer-0.cpp +7 -7
package/src/llama.cpp/tests/test-tokenizer-1-bpe.cpp +5 -5
package/src/llama.cpp/tests/test-tokenizer-1-spm.cpp +5 -5
package/src/llama.cpp/common/train.cpp +0 -1515
package/src/llama.cpp/common/train.h +0 -233
package/src/llama.cpp/examples/baby-llama/CMakeLists.txt +0 -5
package/src/llama.cpp/examples/baby-llama/baby-llama.cpp +0 -1639
package/src/llama.cpp/tests/test-grad0.cpp +0 -1683
/package/src/llama.cpp/ggml/{cmake → src/ggml-cpu/cmake}/FindSIMD.cmake +0 -0
/package/src/llama.cpp/ggml/src/{llamafile → ggml-cpu/llamafile}/sgemm.h +0 -0
/package/src/llama.cpp/ggml/src/{vulkan-shaders → ggml-vulkan/vulkan-shaders}/CMakeLists.txt +0 -0

package/src/llama.cpp/common/arg.cpp CHANGED Viewed

@@ -17,27 +17,27 @@
 using json = nlohmann::ordered_json;
-llama_arg & llama_arg::set_examples(std::initializer_list<enum llama_example> examples) {
+common_arg & common_arg::set_examples(std::initializer_list<enum llama_example> examples) {
     this->examples = std::move(examples);
     return *this;
 }
-llama_arg & llama_arg::set_env(const char * env) {
+common_arg & common_arg::set_env(const char * env) {
     help = help + "\n(env: " + env + ")";
     this->env = env;
     return *this;
 }
-llama_arg & llama_arg::set_sparam() {
+common_arg & common_arg::set_sparam() {
     is_sparam = true;
     return *this;
 }
-bool llama_arg::in_example(enum llama_example ex) {
+bool common_arg::in_example(enum llama_example ex) {
     return examples.find(ex) != examples.end();
 }
-bool llama_arg::get_value_from_env(std::string & output) {
+bool common_arg::get_value_from_env(std::string & output) {
     if (env == nullptr) return false;
     char * value = std::getenv(env);
     if (value) {
@@ -47,7 +47,7 @@ bool llama_arg::get_value_from_env(std::string & output) {
     return false;
 }
-bool llama_arg::has_value_from_env() {
+bool common_arg::has_value_from_env() {
     return env != nullptr && std::getenv(env);
 }
@@ -78,7 +78,7 @@ static std::vector<std::string> break_str_into_lines(std::string input, size_t m
     return result;
 }
-std::string llama_arg::to_string() {
+std::string common_arg::to_string() {
     // params for printing to console
     const static int n_leading_spaces = 40;
     const static int n_char_per_line_help = 70; // TODO: detect this based on current console
@@ -119,33 +119,7 @@ std::string llama_arg::to_string() {
 // utils
 //
-#ifdef __GNUC__
-#ifdef __MINGW32__
-#define LLAMA_COMMON_ATTRIBUTE_FORMAT(...) __attribute__((format(gnu_printf, __VA_ARGS__)))
-#else
-#define LLAMA_COMMON_ATTRIBUTE_FORMAT(...) __attribute__((format(printf, __VA_ARGS__)))
-#endif
-#else
-#define LLAMA_COMMON_ATTRIBUTE_FORMAT(...)
-#endif
-LLAMA_COMMON_ATTRIBUTE_FORMAT(1, 2)
-static std::string format(const char * fmt, ...) {
-    va_list ap;
-    va_list ap2;
-    va_start(ap, fmt);
-    va_copy(ap2, ap);
-    int size = vsnprintf(NULL, 0, fmt, ap);
-    GGML_ASSERT(size >= 0 && size < INT_MAX); // NOLINT
-    std::vector<char> buf(size + 1);
-    int size2 = vsnprintf(buf.data(), size + 1, fmt, ap2);
-    GGML_ASSERT(size2 == size);
-    va_end(ap2);
-    va_end(ap);
-    return std::string(buf.data(), size);
-}
-static void gpt_params_handle_model_default(gpt_params & params) {
+static void common_params_handle_model_default(common_params & params) {
     if (!params.hf_repo.empty()) {
         // short-hand to avoid specifying --hf-file -> default it to --model
         if (params.hf_file.empty()) {
@@ -154,13 +128,13 @@ static void gpt_params_handle_model_default(gpt_params & params) {
             }
             params.hf_file = params.model;
         } else if (params.model.empty()) {
-            params.model = fs_get_cache_file(string_split(params.hf_file, '/').back());
+            params.model = fs_get_cache_file(string_split<std::string>(params.hf_file, '/').back());
         }
     } else if (!params.model_url.empty()) {
         if (params.model.empty()) {
-            auto f = string_split(params.model_url, '#').front();
-            f = string_split(f, '?').front();
-            params.model = fs_get_cache_file(string_split(f, '/').back());
+            auto f = string_split<std::string>(params.model_url, '#').front();
+            f = string_split<std::string>(f, '?').front();
+            params.model = fs_get_cache_file(string_split<std::string>(f, '/').back());
         }
     } else if (params.model.empty()) {
         params.model = DEFAULT_MODEL_PATH;
@@ -171,12 +145,12 @@ static void gpt_params_handle_model_default(gpt_params & params) {
 // CLI argument parsing functions
 //
-static bool gpt_params_parse_ex(int argc, char ** argv, gpt_params_context & ctx_arg) {
+static bool common_params_parse_ex(int argc, char ** argv, common_params_context & ctx_arg) {
     std::string arg;
     const std::string arg_prefix = "--";
-    gpt_params & params = ctx_arg.params;
+    common_params & params = ctx_arg.params;
-    std::unordered_map<std::string, llama_arg *> arg_to_options;
+    std::unordered_map<std::string, common_arg *> arg_to_options;
     for (auto & opt : ctx_arg.options) {
         for (const auto & arg : opt.args) {
             arg_to_options[arg] = &opt;
@@ -199,7 +173,7 @@ static bool gpt_params_parse_ex(int argc, char ** argv, gpt_params_context & ctx
                     continue;
                 }
             } catch (std::exception & e) {
-                throw std::invalid_argument(format(
+                throw std::invalid_argument(string_format(
                     "error while handling environment variable \"%s\": %s\n\n", opt.env, e.what()));
             }
         }
@@ -220,7 +194,7 @@ static bool gpt_params_parse_ex(int argc, char ** argv, gpt_params_context & ctx
             std::replace(arg.begin(), arg.end(), '_', '-');
         }
         if (arg_to_options.find(arg) == arg_to_options.end()) {
-            throw std::invalid_argument(format("error: invalid argument: %s", arg.c_str()));
+            throw std::invalid_argument(string_format("error: invalid argument: %s", arg.c_str()));
         }
         auto opt = *arg_to_options[arg];
         if (opt.has_value_from_env()) {
@@ -252,7 +226,7 @@ static bool gpt_params_parse_ex(int argc, char ** argv, gpt_params_context & ctx
                 continue;
             }
         } catch (std::exception & e) {
-            throw std::invalid_argument(format(
+            throw std::invalid_argument(string_format(
                 "error while handling argument \"%s\": %s\n\n"
                 "usage:\n%s\n\nto show complete usage, run with -h",
                 arg.c_str(), e.what(), arg_to_options[arg]->to_string().c_str()));
@@ -268,7 +242,7 @@ static bool gpt_params_parse_ex(int argc, char ** argv, gpt_params_context & ctx
         throw std::invalid_argument("error: --prompt-cache-all not supported in interactive mode yet\n");
     }
-    gpt_params_handle_model_default(params);
+    common_params_handle_model_default(params);
     if (params.escape) {
         string_process_escapes(params.prompt);
@@ -277,6 +251,9 @@ static bool gpt_params_parse_ex(int argc, char ** argv, gpt_params_context & ctx
         for (auto & antiprompt : params.antiprompt) {
             string_process_escapes(antiprompt);
         }
+        for (auto & seq_breaker : params.sparams.dry_sequence_breakers) {
+            string_process_escapes(seq_breaker);
+        }
     }
     if (!params.kv_overrides.empty()) {
@@ -291,16 +268,16 @@ static bool gpt_params_parse_ex(int argc, char ** argv, gpt_params_context & ctx
     return true;
 }
-static void gpt_params_print_usage(gpt_params_context & ctx_arg) {
-    auto print_options = [](std::vector<llama_arg *> & options) {
-        for (llama_arg * opt : options) {
+static void common_params_print_usage(common_params_context & ctx_arg) {
+    auto print_options = [](std::vector<common_arg *> & options) {
+        for (common_arg * opt : options) {
             printf("%s", opt->to_string().c_str());
         }
     };
-    std::vector<llama_arg *> common_options;
-    std::vector<llama_arg *> sparam_options;
-    std::vector<llama_arg *> specific_options;
+    std::vector<common_arg *> common_options;
+    std::vector<common_arg *> sparam_options;
+    std::vector<common_arg *> specific_options;
     for (auto & opt : ctx_arg.options) {
         // in case multiple LLAMA_EXAMPLE_* are set, we prioritize the LLAMA_EXAMPLE_* matching current example
         if (opt.is_sparam) {
@@ -320,17 +297,17 @@ static void gpt_params_print_usage(gpt_params_context & ctx_arg) {
     print_options(specific_options);
 }
-bool gpt_params_parse(int argc, char ** argv, gpt_params & params, llama_example ex, void(*print_usage)(int, char **)) {
-    auto ctx_arg = gpt_params_parser_init(params, ex, print_usage);
-    const gpt_params params_org = ctx_arg.params; // the example can modify the default params
+bool common_params_parse(int argc, char ** argv, common_params & params, llama_example ex, void(*print_usage)(int, char **)) {
+    auto ctx_arg = common_params_parser_init(params, ex, print_usage);
+    const common_params params_org = ctx_arg.params; // the example can modify the default params
     try {
-        if (!gpt_params_parse_ex(argc, argv, ctx_arg)) {
+        if (!common_params_parse_ex(argc, argv, ctx_arg)) {
             ctx_arg.params = params_org;
             return false;
         }
         if (ctx_arg.params.usage) {
-            gpt_params_print_usage(ctx_arg);
+            common_params_print_usage(ctx_arg);
             if (ctx_arg.print_usage) {
                 ctx_arg.print_usage(argc, argv);
             }
@@ -345,16 +322,16 @@ bool gpt_params_parse(int argc, char ** argv, gpt_params & params, llama_example
     return true;
 }
-gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex, void(*print_usage)(int, char **)) {
-    gpt_params_context ctx_arg(params);
+common_params_context common_params_parser_init(common_params & params, llama_example ex, void(*print_usage)(int, char **)) {
+    common_params_context ctx_arg(params);
     ctx_arg.print_usage = print_usage;
     ctx_arg.ex          = ex;
     std::string sampler_type_chars;
     std::string sampler_type_names;
     for (const auto & sampler : params.sparams.samplers) {
-        sampler_type_chars += gpt_sampler_type_to_chr(sampler);
-        sampler_type_names += gpt_sampler_type_to_str(sampler) + ";";
+        sampler_type_chars += common_sampler_type_to_chr(sampler);
+        sampler_type_names += common_sampler_type_to_str(sampler) + ";";
     }
     sampler_type_names.pop_back();
@@ -366,374 +343,374 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
      * - if LLAMA_EXAMPLE_* is set (other than COMMON), we only show the option in the corresponding example
      * - if both {LLAMA_EXAMPLE_COMMON, LLAMA_EXAMPLE_*,} are set, we will prioritize the LLAMA_EXAMPLE_* matching current example
      */
-    auto add_opt = [&](llama_arg arg) {
+    auto add_opt = [&](common_arg arg) {
         if (arg.in_example(ex) || arg.in_example(LLAMA_EXAMPLE_COMMON)) {
             ctx_arg.options.push_back(std::move(arg));
         }
     };
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-h", "--help", "--usage"},
         "print usage and exit",
-        [](gpt_params & params) {
+        [](common_params & params) {
             params.usage = true;
         }
     ));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--version"},
         "show version and build info",
-        [](gpt_params &) {
+        [](common_params &) {
             fprintf(stderr, "version: %d (%s)\n", LLAMA_BUILD_NUMBER, LLAMA_COMMIT);
             fprintf(stderr, "built with %s for %s\n", LLAMA_COMPILER, LLAMA_BUILD_TARGET);
             exit(0);
         }
     ));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--verbose-prompt"},
-        format("print a verbose prompt before generation (default: %s)", params.verbose_prompt ? "true" : "false"),
-        [](gpt_params & params) {
+        string_format("print a verbose prompt before generation (default: %s)", params.verbose_prompt ? "true" : "false"),
+        [](common_params & params) {
             params.verbose_prompt = true;
         }
     ));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--no-display-prompt"},
-        format("don't print prompt at generation (default: %s)", !params.display_prompt ? "true" : "false"),
-        [](gpt_params & params) {
+        string_format("don't print prompt at generation (default: %s)", !params.display_prompt ? "true" : "false"),
+        [](common_params & params) {
             params.display_prompt = false;
         }
     ).set_examples({LLAMA_EXAMPLE_MAIN}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-co", "--color"},
-        format("colorise output to distinguish prompt and user input from generations (default: %s)", params.use_color ? "true" : "false"),
-        [](gpt_params & params) {
+        string_format("colorise output to distinguish prompt and user input from generations (default: %s)", params.use_color ? "true" : "false"),
+        [](common_params & params) {
             params.use_color = true;
         }
     ).set_examples({LLAMA_EXAMPLE_MAIN, LLAMA_EXAMPLE_INFILL, LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_LOOKUP}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-t", "--threads"}, "N",
-        format("number of threads to use during generation (default: %d)", params.cpuparams.n_threads),
-        [](gpt_params & params, int value) {
+        string_format("number of threads to use during generation (default: %d)", params.cpuparams.n_threads),
+        [](common_params & params, int value) {
             params.cpuparams.n_threads = value;
             if (params.cpuparams.n_threads <= 0) {
                 params.cpuparams.n_threads = std::thread::hardware_concurrency();
             }
         }
     ).set_env("LLAMA_ARG_THREADS"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-tb", "--threads-batch"}, "N",
         "number of threads to use during batch and prompt processing (default: same as --threads)",
-        [](gpt_params & params, int value) {
+        [](common_params & params, int value) {
             params.cpuparams_batch.n_threads = value;
             if (params.cpuparams_batch.n_threads <= 0) {
                 params.cpuparams_batch.n_threads = std::thread::hardware_concurrency();
             }
         }
     ));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-td", "--threads-draft"}, "N",
         "number of threads to use during generation (default: same as --threads)",
-        [](gpt_params & params, int value) {
+        [](common_params & params, int value) {
             params.draft_cpuparams.n_threads = value;
             if (params.draft_cpuparams.n_threads <= 0) {
                 params.draft_cpuparams.n_threads = std::thread::hardware_concurrency();
             }
         }
     ).set_examples({LLAMA_EXAMPLE_SPECULATIVE}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-tbd", "--threads-batch-draft"}, "N",
         "number of threads to use during batch and prompt processing (default: same as --threads-draft)",
-        [](gpt_params & params, int value) {
+        [](common_params & params, int value) {
             params.draft_cpuparams_batch.n_threads = value;
             if (params.draft_cpuparams_batch.n_threads <= 0) {
                 params.draft_cpuparams_batch.n_threads = std::thread::hardware_concurrency();
             }
         }
     ).set_examples({LLAMA_EXAMPLE_SPECULATIVE}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-C", "--cpu-mask"}, "M",
         "CPU affinity mask: arbitrarily long hex. Complements cpu-range (default: \"\")",
-        [](gpt_params & params, const std::string & mask) {
+        [](common_params & params, const std::string & mask) {
             params.cpuparams.mask_valid = true;
             if (!parse_cpu_mask(mask, params.cpuparams.cpumask)) {
                 throw std::invalid_argument("invalid cpumask");
             }
         }
     ));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-Cr", "--cpu-range"}, "lo-hi",
         "range of CPUs for affinity. Complements --cpu-mask",
-        [](gpt_params & params, const std::string & range) {
+        [](common_params & params, const std::string & range) {
             params.cpuparams.mask_valid = true;
             if (!parse_cpu_range(range, params.cpuparams.cpumask)) {
                 throw std::invalid_argument("invalid range");
             }
         }
     ));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--cpu-strict"}, "<0|1>",
-        format("use strict CPU placement (default: %u)\n", (unsigned) params.cpuparams.strict_cpu),
-        [](gpt_params & params, const std::string & value) {
+        string_format("use strict CPU placement (default: %u)\n", (unsigned) params.cpuparams.strict_cpu),
+        [](common_params & params, const std::string & value) {
             params.cpuparams.strict_cpu = std::stoul(value);
         }
     ));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--prio"}, "N",
-        format("set process/thread priority : 0-normal, 1-medium, 2-high, 3-realtime (default: %d)\n", params.cpuparams.priority),
-        [](gpt_params & params, int prio) {
+        string_format("set process/thread priority : 0-normal, 1-medium, 2-high, 3-realtime (default: %d)\n", params.cpuparams.priority),
+        [](common_params & params, int prio) {
             if (prio < 0 || prio > 3) {
                 throw std::invalid_argument("invalid value");
             }
             params.cpuparams.priority = (enum ggml_sched_priority) prio;
         }
     ));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--poll"}, "<0...100>",
-        format("use polling level to wait for work (0 - no polling, default: %u)\n", (unsigned) params.cpuparams.poll),
-        [](gpt_params & params, const std::string & value) {
+        string_format("use polling level to wait for work (0 - no polling, default: %u)\n", (unsigned) params.cpuparams.poll),
+        [](common_params & params, const std::string & value) {
             params.cpuparams.poll = std::stoul(value);
         }
     ));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-Cb", "--cpu-mask-batch"}, "M",
         "CPU affinity mask: arbitrarily long hex. Complements cpu-range-batch (default: same as --cpu-mask)",
-        [](gpt_params & params, const std::string & mask) {
+        [](common_params & params, const std::string & mask) {
             params.cpuparams_batch.mask_valid = true;
             if (!parse_cpu_mask(mask, params.cpuparams_batch.cpumask)) {
                 throw std::invalid_argument("invalid cpumask");
             }
         }
     ));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-Crb", "--cpu-range-batch"}, "lo-hi",
         "ranges of CPUs for affinity. Complements --cpu-mask-batch",
-        [](gpt_params & params, const std::string & range) {
+        [](common_params & params, const std::string & range) {
             params.cpuparams_batch.mask_valid = true;
             if (!parse_cpu_range(range, params.cpuparams_batch.cpumask)) {
                 throw std::invalid_argument("invalid range");
             }
         }
     ));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--cpu-strict-batch"}, "<0|1>",
         "use strict CPU placement (default: same as --cpu-strict)",
-        [](gpt_params & params, int value) {
+        [](common_params & params, int value) {
             params.cpuparams_batch.strict_cpu = value;
         }
     ));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--prio-batch"}, "N",
-        format("set process/thread priority : 0-normal, 1-medium, 2-high, 3-realtime (default: %d)\n", params.cpuparams_batch.priority),
-        [](gpt_params & params, int prio) {
+        string_format("set process/thread priority : 0-normal, 1-medium, 2-high, 3-realtime (default: %d)\n", params.cpuparams_batch.priority),
+        [](common_params & params, int prio) {
             if (prio < 0 || prio > 3) {
                 throw std::invalid_argument("invalid value");
             }
             params.cpuparams_batch.priority = (enum ggml_sched_priority) prio;
         }
     ));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--poll-batch"}, "<0|1>",
         "use polling to wait for work (default: same as --poll)",
-        [](gpt_params & params, int value) {
+        [](common_params & params, int value) {
             params.cpuparams_batch.poll = value;
         }
     ));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-Cd", "--cpu-mask-draft"}, "M",
         "Draft model CPU affinity mask. Complements cpu-range-draft (default: same as --cpu-mask)",
-        [](gpt_params & params, const std::string & mask) {
+        [](common_params & params, const std::string & mask) {
             params.draft_cpuparams.mask_valid = true;
             if (!parse_cpu_mask(mask, params.draft_cpuparams.cpumask)) {
                 throw std::invalid_argument("invalid cpumask");
             }
         }
     ).set_examples({LLAMA_EXAMPLE_SPECULATIVE}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-Crd", "--cpu-range-draft"}, "lo-hi",
         "Ranges of CPUs for affinity. Complements --cpu-mask-draft",
-        [](gpt_params & params, const std::string & range) {
+        [](common_params & params, const std::string & range) {
             params.draft_cpuparams.mask_valid = true;
             if (!parse_cpu_range(range, params.draft_cpuparams.cpumask)) {
                 throw std::invalid_argument("invalid range");
             }
         }
     ).set_examples({LLAMA_EXAMPLE_SPECULATIVE}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--cpu-strict-draft"}, "<0|1>",
         "Use strict CPU placement for draft model (default: same as --cpu-strict)",
-        [](gpt_params & params, int value) {
+        [](common_params & params, int value) {
             params.draft_cpuparams.strict_cpu = value;
         }
     ).set_examples({LLAMA_EXAMPLE_SPECULATIVE}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--prio-draft"}, "N",
-        format("set draft process/thread priority : 0-normal, 1-medium, 2-high, 3-realtime (default: %d)\n", params.draft_cpuparams.priority),
-        [](gpt_params & params, int prio) {
+        string_format("set draft process/thread priority : 0-normal, 1-medium, 2-high, 3-realtime (default: %d)\n", params.draft_cpuparams.priority),
+        [](common_params & params, int prio) {
             if (prio < 0 || prio > 3) {
                 throw std::invalid_argument("invalid value");
             }
             params.draft_cpuparams.priority = (enum ggml_sched_priority) prio;
         }
     ).set_examples({LLAMA_EXAMPLE_SPECULATIVE}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--poll-draft"}, "<0|1>",
         "Use polling to wait for draft model work (default: same as --poll])",
-        [](gpt_params & params, int value) {
+        [](common_params & params, int value) {
             params.draft_cpuparams.poll = value;
         }
     ).set_examples({LLAMA_EXAMPLE_SPECULATIVE}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-Cbd", "--cpu-mask-batch-draft"}, "M",
         "Draft model CPU affinity mask. Complements cpu-range-draft (default: same as --cpu-mask)",
-        [](gpt_params & params, const std::string & mask) {
+        [](common_params & params, const std::string & mask) {
             params.draft_cpuparams_batch.mask_valid = true;
             if (!parse_cpu_mask(mask, params.draft_cpuparams_batch.cpumask)) {
                 throw std::invalid_argument("invalid cpumask");
             }
         }
     ).set_examples({LLAMA_EXAMPLE_SPECULATIVE}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-Crbd", "--cpu-range-batch-draft"}, "lo-hi",
         "Ranges of CPUs for affinity. Complements --cpu-mask-draft-batch)",
-        [](gpt_params & params, const std::string & range) {
+        [](common_params & params, const std::string & range) {
             params.draft_cpuparams_batch.mask_valid = true;
             if (!parse_cpu_range(range, params.draft_cpuparams_batch.cpumask)) {
                 throw std::invalid_argument("invalid cpumask");
             }
         }
     ).set_examples({LLAMA_EXAMPLE_SPECULATIVE}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--cpu-strict-batch-draft"}, "<0|1>",
         "Use strict CPU placement for draft model (default: --cpu-strict-draft)",
-        [](gpt_params & params, int value) {
+        [](common_params & params, int value) {
             params.draft_cpuparams_batch.strict_cpu = value;
         }
     ).set_examples({LLAMA_EXAMPLE_SPECULATIVE}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--prio-batch-draft"}, "N",
-        format("set draft process/thread priority : 0-normal, 1-medium, 2-high, 3-realtime (default: %d)\n", params.draft_cpuparams_batch.priority),
-        [](gpt_params & params, int prio) {
+        string_format("set draft process/thread priority : 0-normal, 1-medium, 2-high, 3-realtime (default: %d)\n", params.draft_cpuparams_batch.priority),
+        [](common_params & params, int prio) {
             if (prio < 0 || prio > 3) {
                 throw std::invalid_argument("invalid value");
             }
             params.draft_cpuparams_batch.priority = (enum ggml_sched_priority) prio;
         }
     ).set_examples({LLAMA_EXAMPLE_SPECULATIVE}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--poll-batch-draft"}, "<0|1>",
         "Use polling to wait for draft model work (default: --poll-draft)",
-        [](gpt_params & params, int value) {
+        [](common_params & params, int value) {
             params.draft_cpuparams_batch.poll = value;
         }
     ).set_examples({LLAMA_EXAMPLE_SPECULATIVE}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--draft"}, "N",
-        format("number of tokens to draft for speculative decoding (default: %d)", params.n_draft),
-        [](gpt_params & params, int value) {
+        string_format("number of tokens to draft for speculative decoding (default: %d)", params.n_draft),
+        [](common_params & params, int value) {
             params.n_draft = value;
         }
     ).set_examples({LLAMA_EXAMPLE_SPECULATIVE, LLAMA_EXAMPLE_LOOKUP}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-ps", "--p-split"}, "N",
-        format("speculative decoding split probability (default: %.1f)", (double)params.p_split),
-        [](gpt_params & params, const std::string & value) {
+        string_format("speculative decoding split probability (default: %.1f)", (double)params.p_split),
+        [](common_params & params, const std::string & value) {
             params.p_split = std::stof(value);
         }
     ).set_examples({LLAMA_EXAMPLE_SPECULATIVE}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-lcs", "--lookup-cache-static"}, "FNAME",
         "path to static lookup cache to use for lookup decoding (not updated by generation)",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             params.lookup_cache_static = value;
         }
     ).set_examples({LLAMA_EXAMPLE_LOOKUP}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-lcd", "--lookup-cache-dynamic"}, "FNAME",
         "path to dynamic lookup cache to use for lookup decoding (updated by generation)",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             params.lookup_cache_dynamic = value;
         }
     ).set_examples({LLAMA_EXAMPLE_LOOKUP}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-c", "--ctx-size"}, "N",
-        format("size of the prompt context (default: %d, 0 = loaded from model)", params.n_ctx),
-        [](gpt_params & params, int value) {
+        string_format("size of the prompt context (default: %d, 0 = loaded from model)", params.n_ctx),
+        [](common_params & params, int value) {
             params.n_ctx = value;
         }
     ).set_env("LLAMA_ARG_CTX_SIZE"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-n", "--predict", "--n-predict"}, "N",
-        format("number of tokens to predict (default: %d, -1 = infinity, -2 = until context filled)", params.n_predict),
-        [](gpt_params & params, int value) {
+        string_format("number of tokens to predict (default: %d, -1 = infinity, -2 = until context filled)", params.n_predict),
+        [](common_params & params, int value) {
             params.n_predict = value;
         }
     ).set_env("LLAMA_ARG_N_PREDICT"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-b", "--batch-size"}, "N",
-        format("logical maximum batch size (default: %d)", params.n_batch),
-        [](gpt_params & params, int value) {
+        string_format("logical maximum batch size (default: %d)", params.n_batch),
+        [](common_params & params, int value) {
             params.n_batch = value;
         }
     ).set_env("LLAMA_ARG_BATCH"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-ub", "--ubatch-size"}, "N",
-        format("physical maximum batch size (default: %d)", params.n_ubatch),
-        [](gpt_params & params, int value) {
+        string_format("physical maximum batch size (default: %d)", params.n_ubatch),
+        [](common_params & params, int value) {
             params.n_ubatch = value;
         }
     ).set_env("LLAMA_ARG_UBATCH"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--keep"}, "N",
-        format("number of tokens to keep from the initial prompt (default: %d, -1 = all)", params.n_keep),
-        [](gpt_params & params, int value) {
+        string_format("number of tokens to keep from the initial prompt (default: %d, -1 = all)", params.n_keep),
+        [](common_params & params, int value) {
             params.n_keep = value;
         }
     ));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--no-context-shift"},
-        format("disables context shift on inifinite text generation (default: %s)", params.ctx_shift ? "disabled" : "enabled"),
-        [](gpt_params & params) {
+        string_format("disables context shift on inifinite text generation (default: %s)", params.ctx_shift ? "disabled" : "enabled"),
+        [](common_params & params) {
             params.ctx_shift = false;
         }
     ).set_examples({LLAMA_EXAMPLE_MAIN, LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_NO_CONTEXT_SHIFT"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--chunks"}, "N",
-        format("max number of chunks to process (default: %d, -1 = all)", params.n_chunks),
-        [](gpt_params & params, int value) {
+        string_format("max number of chunks to process (default: %d, -1 = all)", params.n_chunks),
+        [](common_params & params, int value) {
             params.n_chunks = value;
         }
     ).set_examples({LLAMA_EXAMPLE_IMATRIX, LLAMA_EXAMPLE_PERPLEXITY, LLAMA_EXAMPLE_RETRIEVAL}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-fa", "--flash-attn"},
-        format("enable Flash Attention (default: %s)", params.flash_attn ? "enabled" : "disabled"),
-        [](gpt_params & params) {
+        string_format("enable Flash Attention (default: %s)", params.flash_attn ? "enabled" : "disabled"),
+        [](common_params & params) {
             params.flash_attn = true;
         }
     ).set_env("LLAMA_ARG_FLASH_ATTN"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-p", "--prompt"}, "PROMPT",
         ex == LLAMA_EXAMPLE_MAIN
             ? "prompt to start generation with\nif -cnv is set, this will be used as system prompt"
             : "prompt to start generation with",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             params.prompt = value;
         }
     ));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--no-perf"},
-        format("disable internal libllama performance timings (default: %s)", params.no_perf ? "true" : "false"),
-        [](gpt_params & params) {
+        string_format("disable internal libllama performance timings (default: %s)", params.no_perf ? "true" : "false"),
+        [](common_params & params) {
             params.no_perf = true;
             params.sparams.no_perf = true;
         }
     ).set_env("LLAMA_ARG_NO_PERF"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-f", "--file"}, "FNAME",
         "a file containing the prompt (default: none)",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             std::ifstream file(value);
             if (!file) {
-                throw std::runtime_error(format("error: failed to open file '%s'\n", value.c_str()));
+                throw std::runtime_error(string_format("error: failed to open file '%s'\n", value.c_str()));
             }
             // store the external file name in params
             params.prompt_file = value;
@@ -743,24 +720,24 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
             }
         }
     ));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--in-file"}, "FNAME",
         "an input file (repeat to specify multiple files)",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             std::ifstream file(value);
             if (!file) {
-                throw std::runtime_error(format("error: failed to open file '%s'\n", value.c_str()));
+                throw std::runtime_error(string_format("error: failed to open file '%s'\n", value.c_str()));
             }
             params.in_files.push_back(value);
         }
     ).set_examples({LLAMA_EXAMPLE_IMATRIX}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-bf", "--binary-file"}, "FNAME",
         "binary file containing the prompt (default: none)",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             std::ifstream file(value, std::ios::binary);
             if (!file) {
-                throw std::runtime_error(format("error: failed to open file '%s'\n", value.c_str()));
+                throw std::runtime_error(string_format("error: failed to open file '%s'\n", value.c_str()));
             }
             // store the external file name in params
             params.prompt_file = value;
@@ -770,287 +747,352 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
             fprintf(stderr, "Read %zu bytes from binary file %s\n", params.prompt.size(), value.c_str());
         }
     ));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-e", "--escape"},
-        format("process escapes sequences (\\n, \\r, \\t, \\', \\\", \\\\) (default: %s)", params.escape ? "true" : "false"),
-        [](gpt_params & params) {
+        string_format("process escapes sequences (\\n, \\r, \\t, \\', \\\", \\\\) (default: %s)", params.escape ? "true" : "false"),
+        [](common_params & params) {
             params.escape = true;
         }
     ));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--no-escape"},
         "do not process escape sequences",
-        [](gpt_params & params) {
+        [](common_params & params) {
             params.escape = false;
         }
     ));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-ptc", "--print-token-count"}, "N",
-        format("print token count every N tokens (default: %d)", params.n_print),
-        [](gpt_params & params, int value) {
+        string_format("print token count every N tokens (default: %d)", params.n_print),
+        [](common_params & params, int value) {
             params.n_print = value;
         }
     ).set_examples({LLAMA_EXAMPLE_MAIN}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--prompt-cache"}, "FNAME",
         "file to cache prompt state for faster startup (default: none)",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             params.path_prompt_cache = value;
         }
     ).set_examples({LLAMA_EXAMPLE_MAIN}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--prompt-cache-all"},
         "if specified, saves user input and generations to cache as well\n",
-        [](gpt_params & params) {
+        [](common_params & params) {
             params.prompt_cache_all = true;
         }
     ).set_examples({LLAMA_EXAMPLE_MAIN}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--prompt-cache-ro"},
         "if specified, uses the prompt cache but does not update it",
-        [](gpt_params & params) {
+        [](common_params & params) {
             params.prompt_cache_ro = true;
         }
     ).set_examples({LLAMA_EXAMPLE_MAIN}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-r", "--reverse-prompt"}, "PROMPT",
         "halt generation at PROMPT, return control in interactive mode\n",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             params.antiprompt.emplace_back(value);
         }
     ).set_examples({LLAMA_EXAMPLE_MAIN}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-sp", "--special"},
-        format("special tokens output enabled (default: %s)", params.special ? "true" : "false"),
-        [](gpt_params & params) {
+        string_format("special tokens output enabled (default: %s)", params.special ? "true" : "false"),
+        [](common_params & params) {
             params.special = true;
         }
     ).set_examples({LLAMA_EXAMPLE_MAIN, LLAMA_EXAMPLE_SERVER}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-cnv", "--conversation"},
-        format(
+        string_format(
             "run in conversation mode:\n"
             "- does not print special tokens and suffix/prefix\n"
             "- interactive mode is also enabled\n"
             "(default: %s)",
             params.conversation ? "true" : "false"
         ),
-        [](gpt_params & params) {
+        [](common_params & params) {
             params.conversation = true;
         }
     ).set_examples({LLAMA_EXAMPLE_MAIN}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-i", "--interactive"},
-        format("run in interactive mode (default: %s)", params.interactive ? "true" : "false"),
-        [](gpt_params & params) {
+        string_format("run in interactive mode (default: %s)", params.interactive ? "true" : "false"),
+        [](common_params & params) {
             params.interactive = true;
         }
     ).set_examples({LLAMA_EXAMPLE_MAIN}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-if", "--interactive-first"},
-        format("run in interactive mode and wait for input right away (default: %s)", params.interactive_first ? "true" : "false"),
-        [](gpt_params & params) {
+        string_format("run in interactive mode and wait for input right away (default: %s)", params.interactive_first ? "true" : "false"),
+        [](common_params & params) {
             params.interactive_first = true;
         }
     ).set_examples({LLAMA_EXAMPLE_MAIN}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-mli", "--multiline-input"},
         "allows you to write or paste multiple lines without ending each in '\\'",
-        [](gpt_params & params) {
+        [](common_params & params) {
             params.multiline_input = true;
         }
     ).set_examples({LLAMA_EXAMPLE_MAIN}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--in-prefix-bos"},
         "prefix BOS to user inputs, preceding the `--in-prefix` string",
-        [](gpt_params & params) {
+        [](common_params & params) {
             params.input_prefix_bos = true;
             params.enable_chat_template = false;
         }
     ).set_examples({LLAMA_EXAMPLE_MAIN}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--in-prefix"}, "STRING",
         "string to prefix user inputs with (default: empty)",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             params.input_prefix = value;
             params.enable_chat_template = false;
         }
     ).set_examples({LLAMA_EXAMPLE_MAIN, LLAMA_EXAMPLE_INFILL}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--in-suffix"}, "STRING",
         "string to suffix after user inputs with (default: empty)",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             params.input_suffix = value;
             params.enable_chat_template = false;
         }
     ).set_examples({LLAMA_EXAMPLE_MAIN, LLAMA_EXAMPLE_INFILL}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--no-warmup"},
         "skip warming up the model with an empty run",
-        [](gpt_params & params) {
+        [](common_params & params) {
             params.warmup = false;
         }
     ).set_examples({LLAMA_EXAMPLE_MAIN}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--spm-infill"},
-        format(
+        string_format(
             "use Suffix/Prefix/Middle pattern for infill (instead of Prefix/Suffix/Middle) as some models prefer this. (default: %s)",
             params.spm_infill ? "enabled" : "disabled"
         ),
-        [](gpt_params & params) {
+        [](common_params & params) {
             params.spm_infill = true;
         }
     ).set_examples({LLAMA_EXAMPLE_SERVER, LLAMA_EXAMPLE_INFILL}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--samplers"}, "SAMPLERS",
-        format("samplers that will be used for generation in the order, separated by \';\'\n(default: %s)", sampler_type_names.c_str()),
-        [](gpt_params & params, const std::string & value) {
-            const auto sampler_names = string_split(value, ';');
-            params.sparams.samplers = gpt_sampler_types_from_names(sampler_names, true);
+        string_format("samplers that will be used for generation in the order, separated by \';\'\n(default: %s)", sampler_type_names.c_str()),
+        [](common_params & params, const std::string & value) {
+            const auto sampler_names = string_split<std::string>(value, ';');
+            params.sparams.samplers = common_sampler_types_from_names(sampler_names, true);
         }
     ).set_sparam());
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-s", "--seed"}, "SEED",
-        format("RNG seed (default: %d, use random seed for %d)", params.sparams.seed, LLAMA_DEFAULT_SEED),
-        [](gpt_params & params, const std::string & value) {
+        string_format("RNG seed (default: %d, use random seed for %d)", params.sparams.seed, LLAMA_DEFAULT_SEED),
+        [](common_params & params, const std::string & value) {
             params.sparams.seed = std::stoul(value);
         }
     ).set_sparam());
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--sampling-seq"}, "SEQUENCE",
-        format("simplified sequence for samplers that will be used (default: %s)", sampler_type_chars.c_str()),
-        [](gpt_params & params, const std::string & value) {
-            params.sparams.samplers = gpt_sampler_types_from_chars(value);
+        string_format("simplified sequence for samplers that will be used (default: %s)", sampler_type_chars.c_str()),
+        [](common_params & params, const std::string & value) {
+            params.sparams.samplers = common_sampler_types_from_chars(value);
         }
     ).set_sparam());
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--ignore-eos"},
         "ignore end of stream token and continue generating (implies --logit-bias EOS-inf)",
-        [](gpt_params & params) {
+        [](common_params & params) {
             params.sparams.ignore_eos = true;
         }
     ).set_sparam());
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--penalize-nl"},
-        format("penalize newline tokens (default: %s)", params.sparams.penalize_nl ? "true" : "false"),
-        [](gpt_params & params) {
+        string_format("penalize newline tokens (default: %s)", params.sparams.penalize_nl ? "true" : "false"),
+        [](common_params & params) {
             params.sparams.penalize_nl = true;
         }
     ).set_sparam());
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--temp"}, "N",
-        format("temperature (default: %.1f)", (double)params.sparams.temp),
-        [](gpt_params & params, const std::string & value) {
+        string_format("temperature (default: %.1f)", (double)params.sparams.temp),
+        [](common_params & params, const std::string & value) {
             params.sparams.temp = std::stof(value);
             params.sparams.temp = std::max(params.sparams.temp, 0.0f);
         }
     ).set_sparam());
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--top-k"}, "N",
-        format("top-k sampling (default: %d, 0 = disabled)", params.sparams.top_k),
-        [](gpt_params & params, int value) {
+        string_format("top-k sampling (default: %d, 0 = disabled)", params.sparams.top_k),
+        [](common_params & params, int value) {
             params.sparams.top_k = value;
         }
     ).set_sparam());
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--top-p"}, "N",
-        format("top-p sampling (default: %.1f, 1.0 = disabled)", (double)params.sparams.top_p),
-        [](gpt_params & params, const std::string & value) {
+        string_format("top-p sampling (default: %.1f, 1.0 = disabled)", (double)params.sparams.top_p),
+        [](common_params & params, const std::string & value) {
             params.sparams.top_p = std::stof(value);
         }
     ).set_sparam());
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--min-p"}, "N",
-        format("min-p sampling (default: %.1f, 0.0 = disabled)", (double)params.sparams.min_p),
-        [](gpt_params & params, const std::string & value) {
+        string_format("min-p sampling (default: %.1f, 0.0 = disabled)", (double)params.sparams.min_p),
+        [](common_params & params, const std::string & value) {
             params.sparams.min_p = std::stof(value);
         }
     ).set_sparam());
-    add_opt(llama_arg(
-        {"--tfs"}, "N",
-        format("tail free sampling, parameter z (default: %.1f, 1.0 = disabled)", (double)params.sparams.tfs_z),
-        [](gpt_params & params, const std::string & value) {
-            params.sparams.tfs_z = std::stof(value);
+    add_opt(common_arg(
+        {"--xtc-probability"}, "N",
+        string_format("xtc probability (default: %.1f, 0.0 = disabled)", (double)params.sparams.xtc_probability),
+        [](common_params & params, const std::string & value) {
+            params.sparams.xtc_probability = std::stof(value);
         }
     ).set_sparam());
-    add_opt(llama_arg(
+    add_opt(common_arg(
+        {"--xtc-threshold"}, "N",
+        string_format("xtc threshold (default: %.1f, 1.0 = disabled)", (double)params.sparams.xtc_threshold),
+        [](common_params & params, const std::string & value) {
+            params.sparams.xtc_threshold = std::stof(value);
+        }
+    ).set_sparam());
+    add_opt(common_arg(
         {"--typical"}, "N",
-        format("locally typical sampling, parameter p (default: %.1f, 1.0 = disabled)", (double)params.sparams.typ_p),
-        [](gpt_params & params, const std::string & value) {
+        string_format("locally typical sampling, parameter p (default: %.1f, 1.0 = disabled)", (double)params.sparams.typ_p),
+        [](common_params & params, const std::string & value) {
             params.sparams.typ_p = std::stof(value);
         }
     ).set_sparam());
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--repeat-last-n"}, "N",
-        format("last n tokens to consider for penalize (default: %d, 0 = disabled, -1 = ctx_size)", params.sparams.penalty_last_n),
-        [](gpt_params & params, int value) {
+        string_format("last n tokens to consider for penalize (default: %d, 0 = disabled, -1 = ctx_size)", params.sparams.penalty_last_n),
+        [](common_params & params, int value) {
             params.sparams.penalty_last_n = value;
             params.sparams.n_prev = std::max(params.sparams.n_prev, params.sparams.penalty_last_n);
         }
     ).set_sparam());
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--repeat-penalty"}, "N",
-        format("penalize repeat sequence of tokens (default: %.1f, 1.0 = disabled)", (double)params.sparams.penalty_repeat),
-        [](gpt_params & params, const std::string & value) {
+        string_format("penalize repeat sequence of tokens (default: %.1f, 1.0 = disabled)", (double)params.sparams.penalty_repeat),
+        [](common_params & params, const std::string & value) {
             params.sparams.penalty_repeat = std::stof(value);
         }
     ).set_sparam());
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--presence-penalty"}, "N",
-        format("repeat alpha presence penalty (default: %.1f, 0.0 = disabled)", (double)params.sparams.penalty_present),
-        [](gpt_params & params, const std::string & value) {
+        string_format("repeat alpha presence penalty (default: %.1f, 0.0 = disabled)", (double)params.sparams.penalty_present),
+        [](common_params & params, const std::string & value) {
             params.sparams.penalty_present = std::stof(value);
         }
     ).set_sparam());
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--frequency-penalty"}, "N",
-        format("repeat alpha frequency penalty (default: %.1f, 0.0 = disabled)", (double)params.sparams.penalty_freq),
-        [](gpt_params & params, const std::string & value) {
+        string_format("repeat alpha frequency penalty (default: %.1f, 0.0 = disabled)", (double)params.sparams.penalty_freq),
+        [](common_params & params, const std::string & value) {
             params.sparams.penalty_freq = std::stof(value);
         }
     ).set_sparam());
-    add_opt(llama_arg(
+    add_opt(common_arg(
+        {"--dry-multiplier"}, "N",
+        string_format("set DRY sampling multiplier (default: %.1f, 0.0 = disabled)", (double)params.sparams.dry_multiplier),
+        [](common_params & params, const std::string & value) {
+            params.sparams.dry_multiplier = std::stof(value);
+        }
+    ).set_sparam());
+    add_opt(common_arg(
+        {"--dry-base"}, "N",
+        string_format("set DRY sampling base value (default: %.2f)", (double)params.sparams.dry_base),
+        [](common_params & params, const std::string & value) {
+            float potential_base = std::stof(value);
+            if (potential_base >= 1.0f)
+            {
+                params.sparams.dry_base = potential_base;
+            }
+        }
+    ).set_sparam());
+    add_opt(common_arg(
+        {"--dry-allowed-length"}, "N",
+        string_format("set allowed length for DRY sampling (default: %d)", params.sparams.dry_allowed_length),
+        [](common_params & params, int value) {
+            params.sparams.dry_allowed_length = value;
+        }
+    ).set_sparam());
+    add_opt(common_arg(
+        {"--dry-penalty-last-n"}, "N",
+        string_format("set DRY penalty for the last n tokens (default: %d, 0 = disable, -1 = context size)", params.sparams.dry_penalty_last_n),
+        [](common_params & params, int value) {
+            params.sparams.dry_penalty_last_n = value;
+        }
+    ).set_sparam());
+    add_opt(common_arg(
+        {"--dry-sequence-breaker"}, "STRING",
+        string_format("add sequence breaker for DRY sampling, clearing out default breakers (%s) in the process; use \"none\" to not use any sequence breakers\n",
+            params.sparams.dry_sequence_breakers.empty() ? "none" :
+            std::accumulate(std::next(params.sparams.dry_sequence_breakers.begin()),
+                params.sparams.dry_sequence_breakers.end(),
+                std::string("'") + (params.sparams.dry_sequence_breakers[0] == "\n" ? "\\n" : params.sparams.dry_sequence_breakers[0]) + "'",
+                [](const std::string& a, const std::string& b) {
+                    std::string formatted_b = (b == "\n") ? "\\n" : b;
+                    return a + ", '" + formatted_b + "'";
+                }).c_str()),
+        [](common_params & params, const std::string & value) {
+            static bool defaults_cleared = false;
+            if (!defaults_cleared) {
+                params.sparams.dry_sequence_breakers.clear();
+                defaults_cleared = true;
+            }
+            if (value == "none") {
+                params.sparams.dry_sequence_breakers.clear();
+            } else {
+                params.sparams.dry_sequence_breakers.emplace_back(value);
+            }
+        }
+    ).set_sparam());
+    add_opt(common_arg(
         {"--dynatemp-range"}, "N",
-        format("dynamic temperature range (default: %.1f, 0.0 = disabled)", (double)params.sparams.dynatemp_range),
-        [](gpt_params & params, const std::string & value) {
+        string_format("dynamic temperature range (default: %.1f, 0.0 = disabled)", (double)params.sparams.dynatemp_range),
+        [](common_params & params, const std::string & value) {
             params.sparams.dynatemp_range = std::stof(value);
         }
     ).set_sparam());
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--dynatemp-exp"}, "N",
-        format("dynamic temperature exponent (default: %.1f)", (double)params.sparams.dynatemp_exponent),
-        [](gpt_params & params, const std::string & value) {
+        string_format("dynamic temperature exponent (default: %.1f)", (double)params.sparams.dynatemp_exponent),
+        [](common_params & params, const std::string & value) {
             params.sparams.dynatemp_exponent = std::stof(value);
         }
     ).set_sparam());
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--mirostat"}, "N",
-        format("use Mirostat sampling.\nTop K, Nucleus, Tail Free and Locally Typical samplers are ignored if used.\n"
+        string_format("use Mirostat sampling.\nTop K, Nucleus and Locally Typical samplers are ignored if used.\n"
         "(default: %d, 0 = disabled, 1 = Mirostat, 2 = Mirostat 2.0)", params.sparams.mirostat),
-        [](gpt_params & params, int value) {
+        [](common_params & params, int value) {
             params.sparams.mirostat = value;
         }
     ).set_sparam());
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--mirostat-lr"}, "N",
-        format("Mirostat learning rate, parameter eta (default: %.1f)", (double)params.sparams.mirostat_eta),
-        [](gpt_params & params, const std::string & value) {
+        string_format("Mirostat learning rate, parameter eta (default: %.1f)", (double)params.sparams.mirostat_eta),
+        [](common_params & params, const std::string & value) {
             params.sparams.mirostat_eta = std::stof(value);
         }
     ).set_sparam());
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--mirostat-ent"}, "N",
-        format("Mirostat target entropy, parameter tau (default: %.1f)", (double)params.sparams.mirostat_tau),
-        [](gpt_params & params, const std::string & value) {
+        string_format("Mirostat target entropy, parameter tau (default: %.1f)", (double)params.sparams.mirostat_tau),
+        [](common_params & params, const std::string & value) {
             params.sparams.mirostat_tau = std::stof(value);
         }
     ).set_sparam());
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-l", "--logit-bias"}, "TOKEN_ID(+/-)BIAS",
         "modifies the likelihood of token appearing in the completion,\n"
         "i.e. `--logit-bias 15043+1` to increase likelihood of token ' Hello',\n"
         "or `--logit-bias 15043-1` to decrease likelihood of token ' Hello'",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             std::stringstream ss(value);
             llama_token key;
             char sign;
@@ -1067,20 +1109,20 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
             }
         }
     ).set_sparam());
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--grammar"}, "GRAMMAR",
-        format("BNF-like grammar to constrain generations (see samples in grammars/ dir) (default: '%s')", params.sparams.grammar.c_str()),
-        [](gpt_params & params, const std::string & value) {
+        string_format("BNF-like grammar to constrain generations (see samples in grammars/ dir) (default: '%s')", params.sparams.grammar.c_str()),
+        [](common_params & params, const std::string & value) {
             params.sparams.grammar = value;
         }
     ).set_sparam());
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--grammar-file"}, "FNAME",
         "file to read grammar from",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             std::ifstream file(value);
             if (!file) {
-                throw std::runtime_error(format("error: failed to open file '%s'\n", value.c_str()));
+                throw std::runtime_error(string_format("error: failed to open file '%s'\n", value.c_str()));
             }
             std::copy(
                 std::istreambuf_iterator<char>(file),
@@ -1089,17 +1131,17 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
             );
         }
     ).set_sparam());
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-j", "--json-schema"}, "SCHEMA",
         "JSON schema to constrain generations (https://json-schema.org/), e.g. `{}` for any JSON object\nFor schemas w/ external $refs, use --grammar + example/json_schema_to_grammar.py instead",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             params.sparams.grammar = json_schema_to_grammar(json::parse(value));
         }
     ).set_sparam());
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--pooling"}, "{none,mean,cls,last,rank}",
         "pooling type for embeddings, use model default if unspecified",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             /**/ if (value == "none") { params.pooling_type = LLAMA_POOLING_TYPE_NONE; }
             else if (value == "mean") { params.pooling_type = LLAMA_POOLING_TYPE_MEAN; }
             else if (value == "cls")  { params.pooling_type = LLAMA_POOLING_TYPE_CLS;  }
@@ -1108,275 +1150,275 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
             else { throw std::invalid_argument("invalid value"); }
         }
     ).set_examples({LLAMA_EXAMPLE_EMBEDDING, LLAMA_EXAMPLE_RETRIEVAL, LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_POOLING"));
-    add_opt(llama_arg(
-        {"--attention"}, "{causal,non,causal}",
+    add_opt(common_arg(
+        {"--attention"}, "{causal,non-causal}",
         "attention type for embeddings, use model default if unspecified",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             /**/ if (value == "causal") { params.attention_type = LLAMA_ATTENTION_TYPE_CAUSAL; }
             else if (value == "non-causal") { params.attention_type = LLAMA_ATTENTION_TYPE_NON_CAUSAL; }
             else { throw std::invalid_argument("invalid value"); }
         }
     ).set_examples({LLAMA_EXAMPLE_EMBEDDING}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--rope-scaling"}, "{none,linear,yarn}",
         "RoPE frequency scaling method, defaults to linear unless specified by the model",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             /**/ if (value == "none") { params.rope_scaling_type = LLAMA_ROPE_SCALING_TYPE_NONE; }
             else if (value == "linear") { params.rope_scaling_type = LLAMA_ROPE_SCALING_TYPE_LINEAR; }
             else if (value == "yarn") { params.rope_scaling_type = LLAMA_ROPE_SCALING_TYPE_YARN; }
             else { throw std::invalid_argument("invalid value"); }
         }
     ).set_env("LLAMA_ARG_ROPE_SCALING_TYPE"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--rope-scale"}, "N",
         "RoPE context scaling factor, expands context by a factor of N",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             params.rope_freq_scale = 1.0f / std::stof(value);
         }
     ).set_env("LLAMA_ARG_ROPE_SCALE"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--rope-freq-base"}, "N",
         "RoPE base frequency, used by NTK-aware scaling (default: loaded from model)",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             params.rope_freq_base = std::stof(value);
         }
     ).set_env("LLAMA_ARG_ROPE_FREQ_BASE"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--rope-freq-scale"}, "N",
         "RoPE frequency scaling factor, expands context by a factor of 1/N",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             params.rope_freq_scale = std::stof(value);
         }
     ).set_env("LLAMA_ARG_ROPE_FREQ_SCALE"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--yarn-orig-ctx"}, "N",
-        format("YaRN: original context size of model (default: %d = model training context size)", params.yarn_orig_ctx),
-        [](gpt_params & params, int value) {
+        string_format("YaRN: original context size of model (default: %d = model training context size)", params.yarn_orig_ctx),
+        [](common_params & params, int value) {
             params.yarn_orig_ctx = value;
         }
     ).set_env("LLAMA_ARG_YARN_ORIG_CTX"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--yarn-ext-factor"}, "N",
-        format("YaRN: extrapolation mix factor (default: %.1f, 0.0 = full interpolation)", (double)params.yarn_ext_factor),
-        [](gpt_params & params, const std::string & value) {
+        string_format("YaRN: extrapolation mix factor (default: %.1f, 0.0 = full interpolation)", (double)params.yarn_ext_factor),
+        [](common_params & params, const std::string & value) {
             params.yarn_ext_factor = std::stof(value);
         }
     ).set_env("LLAMA_ARG_YARN_EXT_FACTOR"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--yarn-attn-factor"}, "N",
-        format("YaRN: scale sqrt(t) or attention magnitude (default: %.1f)", (double)params.yarn_attn_factor),
-        [](gpt_params & params, const std::string & value) {
+        string_format("YaRN: scale sqrt(t) or attention magnitude (default: %.1f)", (double)params.yarn_attn_factor),
+        [](common_params & params, const std::string & value) {
             params.yarn_attn_factor = std::stof(value);
         }
     ).set_env("LLAMA_ARG_YARN_ATTN_FACTOR"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--yarn-beta-slow"}, "N",
-        format("YaRN: high correction dim or alpha (default: %.1f)", (double)params.yarn_beta_slow),
-        [](gpt_params & params, const std::string & value) {
+        string_format("YaRN: high correction dim or alpha (default: %.1f)", (double)params.yarn_beta_slow),
+        [](common_params & params, const std::string & value) {
             params.yarn_beta_slow = std::stof(value);
         }
     ).set_env("LLAMA_ARG_YARN_BETA_SLOW"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--yarn-beta-fast"}, "N",
-        format("YaRN: low correction dim or beta (default: %.1f)", (double)params.yarn_beta_fast),
-        [](gpt_params & params, const std::string & value) {
+        string_format("YaRN: low correction dim or beta (default: %.1f)", (double)params.yarn_beta_fast),
+        [](common_params & params, const std::string & value) {
             params.yarn_beta_fast = std::stof(value);
         }
     ).set_env("LLAMA_ARG_YARN_BETA_FAST"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-gan", "--grp-attn-n"}, "N",
-        format("group-attention factor (default: %d)", params.grp_attn_n),
-        [](gpt_params & params, int value) {
+        string_format("group-attention factor (default: %d)", params.grp_attn_n),
+        [](common_params & params, int value) {
             params.grp_attn_n = value;
         }
-    ).set_env("LLAMA_ARG_GRP_ATTN_N"));
-    add_opt(llama_arg(
+    ).set_env("LLAMA_ARG_GRP_ATTN_N").set_examples({LLAMA_EXAMPLE_MAIN, LLAMA_EXAMPLE_PASSKEY}));
+    add_opt(common_arg(
         {"-gaw", "--grp-attn-w"}, "N",
-        format("group-attention width (default: %.1f)", (double)params.grp_attn_w),
-        [](gpt_params & params, int value) {
+        string_format("group-attention width (default: %d)", params.grp_attn_w),
+        [](common_params & params, int value) {
             params.grp_attn_w = value;
         }
-    ).set_env("LLAMA_ARG_GRP_ATTN_W"));
-    add_opt(llama_arg(
+    ).set_env("LLAMA_ARG_GRP_ATTN_W").set_examples({LLAMA_EXAMPLE_MAIN}));
+    add_opt(common_arg(
         {"-dkvc", "--dump-kv-cache"},
         "verbose print of the KV cache",
-        [](gpt_params & params) {
+        [](common_params & params) {
             params.dump_kv_cache = true;
         }
     ));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-nkvo", "--no-kv-offload"},
         "disable KV offload",
-        [](gpt_params & params) {
+        [](common_params & params) {
             params.no_kv_offload = true;
         }
     ).set_env("LLAMA_ARG_NO_KV_OFFLOAD"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-ctk", "--cache-type-k"}, "TYPE",
-        format("KV cache data type for K (default: %s)", params.cache_type_k.c_str()),
-        [](gpt_params & params, const std::string & value) {
+        string_format("KV cache data type for K (default: %s)", params.cache_type_k.c_str()),
+        [](common_params & params, const std::string & value) {
             // TODO: get the type right here
             params.cache_type_k = value;
         }
     ).set_env("LLAMA_ARG_CACHE_TYPE_K"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-ctv", "--cache-type-v"}, "TYPE",
-        format("KV cache data type for V (default: %s)", params.cache_type_v.c_str()),
-        [](gpt_params & params, const std::string & value) {
+        string_format("KV cache data type for V (default: %s)", params.cache_type_v.c_str()),
+        [](common_params & params, const std::string & value) {
             // TODO: get the type right here
             params.cache_type_v = value;
         }
     ).set_env("LLAMA_ARG_CACHE_TYPE_V"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--perplexity", "--all-logits"},
-        format("return logits for all tokens in the batch (default: %s)", params.logits_all ? "true" : "false"),
-        [](gpt_params & params) {
+        string_format("return logits for all tokens in the batch (default: %s)", params.logits_all ? "true" : "false"),
+        [](common_params & params) {
             params.logits_all = true;
         }
     ).set_examples({LLAMA_EXAMPLE_PERPLEXITY}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--hellaswag"},
         "compute HellaSwag score over random tasks from datafile supplied with -f",
-        [](gpt_params & params) {
+        [](common_params & params) {
             params.hellaswag = true;
         }
     ).set_examples({LLAMA_EXAMPLE_PERPLEXITY}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--hellaswag-tasks"}, "N",
-        format("number of tasks to use when computing the HellaSwag score (default: %zu)", params.hellaswag_tasks),
-        [](gpt_params & params, int value) {
+        string_format("number of tasks to use when computing the HellaSwag score (default: %zu)", params.hellaswag_tasks),
+        [](common_params & params, int value) {
             params.hellaswag_tasks = value;
         }
     ).set_examples({LLAMA_EXAMPLE_PERPLEXITY}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--winogrande"},
         "compute Winogrande score over random tasks from datafile supplied with -f",
-        [](gpt_params & params) {
+        [](common_params & params) {
             params.winogrande = true;
         }
     ).set_examples({LLAMA_EXAMPLE_PERPLEXITY}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--winogrande-tasks"}, "N",
-        format("number of tasks to use when computing the Winogrande score (default: %zu)", params.winogrande_tasks),
-        [](gpt_params & params, int value) {
+        string_format("number of tasks to use when computing the Winogrande score (default: %zu)", params.winogrande_tasks),
+        [](common_params & params, int value) {
             params.winogrande_tasks = value;
         }
     ).set_examples({LLAMA_EXAMPLE_PERPLEXITY}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--multiple-choice"},
         "compute multiple choice score over random tasks from datafile supplied with -f",
-        [](gpt_params & params) {
+        [](common_params & params) {
             params.multiple_choice = true;
         }
     ).set_examples({LLAMA_EXAMPLE_PERPLEXITY}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--multiple-choice-tasks"}, "N",
-        format("number of tasks to use when computing the multiple choice score (default: %zu)", params.multiple_choice_tasks),
-        [](gpt_params & params, int value) {
+        string_format("number of tasks to use when computing the multiple choice score (default: %zu)", params.multiple_choice_tasks),
+        [](common_params & params, int value) {
             params.multiple_choice_tasks = value;
         }
     ).set_examples({LLAMA_EXAMPLE_PERPLEXITY}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--kl-divergence"},
         "computes KL-divergence to logits provided via --kl-divergence-base",
-        [](gpt_params & params) {
+        [](common_params & params) {
             params.kl_divergence = true;
         }
     ).set_examples({LLAMA_EXAMPLE_PERPLEXITY}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--save-all-logits", "--kl-divergence-base"}, "FNAME",
         "set logits file",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             params.logits_file = value;
         }
     ).set_examples({LLAMA_EXAMPLE_PERPLEXITY}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--ppl-stride"}, "N",
-        format("stride for perplexity calculation (default: %d)", params.ppl_stride),
-        [](gpt_params & params, int value) {
+        string_format("stride for perplexity calculation (default: %d)", params.ppl_stride),
+        [](common_params & params, int value) {
             params.ppl_stride = value;
         }
     ).set_examples({LLAMA_EXAMPLE_PERPLEXITY}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--ppl-output-type"}, "<0|1>",
-        format("output type for perplexity calculation (default: %d)", params.ppl_output_type),
-        [](gpt_params & params, int value) {
+        string_format("output type for perplexity calculation (default: %d)", params.ppl_output_type),
+        [](common_params & params, int value) {
             params.ppl_output_type = value;
         }
     ).set_examples({LLAMA_EXAMPLE_PERPLEXITY}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-dt", "--defrag-thold"}, "N",
-        format("KV cache defragmentation threshold (default: %.1f, < 0 - disabled)", (double)params.defrag_thold),
-        [](gpt_params & params, const std::string & value) {
+        string_format("KV cache defragmentation threshold (default: %.1f, < 0 - disabled)", (double)params.defrag_thold),
+        [](common_params & params, const std::string & value) {
             params.defrag_thold = std::stof(value);
         }
     ).set_env("LLAMA_ARG_DEFRAG_THOLD"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-np", "--parallel"}, "N",
-        format("number of parallel sequences to decode (default: %d)", params.n_parallel),
-        [](gpt_params & params, int value) {
+        string_format("number of parallel sequences to decode (default: %d)", params.n_parallel),
+        [](common_params & params, int value) {
             params.n_parallel = value;
         }
     ).set_env("LLAMA_ARG_N_PARALLEL"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-ns", "--sequences"}, "N",
-        format("number of sequences to decode (default: %d)", params.n_sequences),
-        [](gpt_params & params, int value) {
+        string_format("number of sequences to decode (default: %d)", params.n_sequences),
+        [](common_params & params, int value) {
             params.n_sequences = value;
         }
     ).set_examples({LLAMA_EXAMPLE_PARALLEL}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-cb", "--cont-batching"},
-        format("enable continuous batching (a.k.a dynamic batching) (default: %s)", params.cont_batching ? "enabled" : "disabled"),
-        [](gpt_params & params) {
+        string_format("enable continuous batching (a.k.a dynamic batching) (default: %s)", params.cont_batching ? "enabled" : "disabled"),
+        [](common_params & params) {
             params.cont_batching = true;
         }
     ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_CONT_BATCHING"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-nocb", "--no-cont-batching"},
         "disable continuous batching",
-        [](gpt_params & params) {
+        [](common_params & params) {
             params.cont_batching = false;
         }
     ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_NO_CONT_BATCHING"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--mmproj"}, "FILE",
         "path to a multimodal projector file for LLaVA. see examples/llava/README.md",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             params.mmproj = value;
         }
     ).set_examples({LLAMA_EXAMPLE_LLAVA}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--image"}, "FILE",
         "path to an image file. use with multimodal models. Specify multiple times for batching",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             params.image.emplace_back(value);
         }
     ).set_examples({LLAMA_EXAMPLE_LLAVA}));
-#ifdef GGML_USE_RPC
-    add_opt(llama_arg(
-        {"--rpc"}, "SERVERS",
-        "comma separated list of RPC servers",
-        [](gpt_params & params, const std::string & value) {
-            params.rpc_servers = value;
-        }
-    ).set_env("LLAMA_ARG_RPC"));
-#endif
-    add_opt(llama_arg(
+    if (llama_supports_rpc()) {
+        add_opt(common_arg(
+            {"--rpc"}, "SERVERS",
+            "comma separated list of RPC servers",
+            [](common_params & params, const std::string & value) {
+                params.rpc_servers = value;
+            }
+        ).set_env("LLAMA_ARG_RPC"));
+    }
+    add_opt(common_arg(
         {"--mlock"},
         "force system to keep model in RAM rather than swapping or compressing",
-        [](gpt_params & params) {
+        [](common_params & params) {
             params.use_mlock = true;
         }
     ).set_env("LLAMA_ARG_MLOCK"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--no-mmap"},
         "do not memory-map model (slower load but may reduce pageouts if not using mlock)",
-        [](gpt_params & params) {
+        [](common_params & params) {
             params.use_mmap = false;
         }
     ).set_env("LLAMA_ARG_NO_MMAP"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--numa"}, "TYPE",
         "attempt optimizations that help on some NUMA systems\n"
         "- distribute: spread execution evenly over all nodes\n"
@@ -1384,17 +1426,17 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
         "- numactl: use the CPU map provided by numactl\n"
         "if run without this previously, it is recommended to drop the system page cache before using this\n"
         "see https://github.com/ggerganov/llama.cpp/issues/1437",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             /**/ if (value == "distribute" || value == "") { params.numa = GGML_NUMA_STRATEGY_DISTRIBUTE; }
             else if (value == "isolate") { params.numa = GGML_NUMA_STRATEGY_ISOLATE; }
             else if (value == "numactl") { params.numa = GGML_NUMA_STRATEGY_NUMACTL; }
             else { throw std::invalid_argument("invalid value"); }
         }
     ).set_env("LLAMA_ARG_NUMA"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-ngl", "--gpu-layers", "--n-gpu-layers"}, "N",
         "number of layers to store in VRAM",
-        [](gpt_params & params, int value) {
+        [](common_params & params, int value) {
             params.n_gpu_layers = value;
             if (!llama_supports_gpu_offload()) {
                 fprintf(stderr, "warning: not compiled with GPU offload support, --gpu-layers option will be ignored\n");
@@ -1402,10 +1444,10 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
             }
         }
     ).set_env("LLAMA_ARG_N_GPU_LAYERS"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-ngld", "--gpu-layers-draft", "--n-gpu-layers-draft"}, "N",
         "number of layers to store in VRAM for the draft model",
-        [](gpt_params & params, int value) {
+        [](common_params & params, int value) {
             params.n_gpu_layers_draft = value;
             if (!llama_supports_gpu_offload()) {
                 fprintf(stderr, "warning: not compiled with GPU offload support, --gpu-layers-draft option will be ignored\n");
@@ -1413,13 +1455,13 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
             }
         }
     ).set_examples({LLAMA_EXAMPLE_SPECULATIVE}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-sm", "--split-mode"}, "{none,layer,row}",
         "how to split the model across multiple GPUs, one of:\n"
         "- none: use one GPU only\n"
         "- layer (default): split layers and KV across GPUs\n"
         "- row: split rows across GPUs",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             std::string arg_next = value;
             if (arg_next == "none") {
                 params.split_mode = LLAMA_SPLIT_MODE_NONE;
@@ -1439,10 +1481,10 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
             }
         }
     ).set_env("LLAMA_ARG_SPLIT_MODE"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-ts", "--tensor-split"}, "N0,N1,N2,...",
         "fraction of the model to offload to each GPU, comma-separated list of proportions, e.g. 3,1",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             std::string arg_next = value;
             // split string by , and /
@@ -1451,7 +1493,7 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
             std::vector<std::string> split_arg{ it, {} };
             if (split_arg.size() >= llama_max_devices()) {
                 throw std::invalid_argument(
-                    format("got %d input configs, but system only has %d devices", (int)split_arg.size(), (int)llama_max_devices())
+                    string_format("got %d input configs, but system only has %d devices", (int)split_arg.size(), (int)llama_max_devices())
                 );
             }
             for (size_t i = 0; i < llama_max_devices(); ++i) {
@@ -1466,315 +1508,315 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
             }
         }
     ).set_env("LLAMA_ARG_TENSOR_SPLIT"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-mg", "--main-gpu"}, "INDEX",
-        format("the GPU to use for the model (with split-mode = none), or for intermediate results and KV (with split-mode = row) (default: %d)", params.main_gpu),
-        [](gpt_params & params, int value) {
+        string_format("the GPU to use for the model (with split-mode = none), or for intermediate results and KV (with split-mode = row) (default: %d)", params.main_gpu),
+        [](common_params & params, int value) {
             params.main_gpu = value;
             if (!llama_supports_gpu_offload()) {
                 fprintf(stderr, "warning: llama.cpp was compiled without support for GPU offload. Setting the main GPU has no effect.\n");
             }
         }
     ).set_env("LLAMA_ARG_MAIN_GPU"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--check-tensors"},
-        format("check model tensor data for invalid values (default: %s)", params.check_tensors ? "true" : "false"),
-        [](gpt_params & params) {
+        string_format("check model tensor data for invalid values (default: %s)", params.check_tensors ? "true" : "false"),
+        [](common_params & params) {
             params.check_tensors = true;
         }
     ));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--override-kv"}, "KEY=TYPE:VALUE",
         "advanced option to override model metadata by key. may be specified multiple times.\n"
         "types: int, float, bool, str. example: --override-kv tokenizer.ggml.add_bos_token=bool:false",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             if (!string_parse_kv_override(value.c_str(), params.kv_overrides)) {
-                throw std::runtime_error(format("error: Invalid type for KV override: %s\n", value.c_str()));
+                throw std::runtime_error(string_format("error: Invalid type for KV override: %s\n", value.c_str()));
             }
         }
     ));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--lora"}, "FNAME",
         "path to LoRA adapter (can be repeated to use multiple adapters)",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             params.lora_adapters.push_back({ std::string(value), 1.0 });
         }
         // we define this arg on both COMMON and EXPORT_LORA, so when showing help message of export-lora, it will be categorized as "example-specific" arg
     ).set_examples({LLAMA_EXAMPLE_COMMON, LLAMA_EXAMPLE_EXPORT_LORA}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--lora-scaled"}, "FNAME", "SCALE",
         "path to LoRA adapter with user defined scaling (can be repeated to use multiple adapters)",
-        [](gpt_params & params, const std::string & fname, const std::string & scale) {
+        [](common_params & params, const std::string & fname, const std::string & scale) {
             params.lora_adapters.push_back({ fname, std::stof(scale) });
         }
         // we define this arg on both COMMON and EXPORT_LORA, so when showing help message of export-lora, it will be categorized as "example-specific" arg
     ).set_examples({LLAMA_EXAMPLE_COMMON, LLAMA_EXAMPLE_EXPORT_LORA}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--control-vector"}, "FNAME",
         "add a control vector\nnote: this argument can be repeated to add multiple control vectors",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             params.control_vectors.push_back({ 1.0f, value, });
         }
     ));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--control-vector-scaled"}, "FNAME", "SCALE",
         "add a control vector with user defined scaling SCALE\n"
         "note: this argument can be repeated to add multiple scaled control vectors",
-        [](gpt_params & params, const std::string & fname, const std::string & scale) {
+        [](common_params & params, const std::string & fname, const std::string & scale) {
             params.control_vectors.push_back({ std::stof(scale), fname });
         }
     ));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--control-vector-layer-range"}, "START", "END",
         "layer range to apply the control vector(s) to, start and end inclusive",
-        [](gpt_params & params, const std::string & start, const std::string & end) {
+        [](common_params & params, const std::string & start, const std::string & end) {
             params.control_vector_layer_start = std::stoi(start);
             params.control_vector_layer_end = std::stoi(end);
         }
     ));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-a", "--alias"}, "STRING",
         "set alias for model name (to be used by REST API)",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             params.model_alias = value;
         }
     ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_ALIAS"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-m", "--model"}, "FNAME",
         ex == LLAMA_EXAMPLE_EXPORT_LORA
             ? std::string("model path from which to load base model")
-            : format(
+            : string_format(
                 "model path (default: `models/$filename` with filename from `--hf-file` "
                 "or `--model-url` if set, otherwise %s)", DEFAULT_MODEL_PATH
             ),
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             params.model = value;
         }
     ).set_examples({LLAMA_EXAMPLE_COMMON, LLAMA_EXAMPLE_EXPORT_LORA}).set_env("LLAMA_ARG_MODEL"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-md", "--model-draft"}, "FNAME",
         "draft model for speculative decoding (default: unused)",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             params.model_draft = value;
         }
     ).set_examples({LLAMA_EXAMPLE_SPECULATIVE}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-mu", "--model-url"}, "MODEL_URL",
         "model download url (default: unused)",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             params.model_url = value;
         }
     ).set_env("LLAMA_ARG_MODEL_URL"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-hfr", "--hf-repo"}, "REPO",
         "Hugging Face model repository (default: unused)",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             params.hf_repo = value;
         }
     ).set_env("LLAMA_ARG_HF_REPO"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-hff", "--hf-file"}, "FILE",
         "Hugging Face model file (default: unused)",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             params.hf_file = value;
         }
     ).set_env("LLAMA_ARG_HF_FILE"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-hft", "--hf-token"}, "TOKEN",
         "Hugging Face access token (default: value from HF_TOKEN environment variable)",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             params.hf_token = value;
         }
     ).set_env("HF_TOKEN"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--context-file"}, "FNAME",
         "file to load context from (repeat to specify multiple files)",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             std::ifstream file(value, std::ios::binary);
             if (!file) {
-                throw std::runtime_error(format("error: failed to open file '%s'\n", value.c_str()));
+                throw std::runtime_error(string_format("error: failed to open file '%s'\n", value.c_str()));
             }
             params.context_files.push_back(value);
         }
     ).set_examples({LLAMA_EXAMPLE_RETRIEVAL}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--chunk-size"}, "N",
-        format("minimum length of embedded text chunks (default: %d)", params.chunk_size),
-        [](gpt_params & params, int value) {
+        string_format("minimum length of embedded text chunks (default: %d)", params.chunk_size),
+        [](common_params & params, int value) {
             params.chunk_size = value;
         }
     ).set_examples({LLAMA_EXAMPLE_RETRIEVAL}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--chunk-separator"}, "STRING",
-        format("separator between chunks (default: '%s')", params.chunk_separator.c_str()),
-        [](gpt_params & params, const std::string & value) {
+        string_format("separator between chunks (default: '%s')", params.chunk_separator.c_str()),
+        [](common_params & params, const std::string & value) {
             params.chunk_separator = value;
         }
     ).set_examples({LLAMA_EXAMPLE_RETRIEVAL}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--junk"}, "N",
-        format("number of times to repeat the junk text (default: %d)", params.n_junk),
-        [](gpt_params & params, int value) {
+        string_format("number of times to repeat the junk text (default: %d)", params.n_junk),
+        [](common_params & params, int value) {
             params.n_junk = value;
         }
     ).set_examples({LLAMA_EXAMPLE_PASSKEY}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--pos"}, "N",
-        format("position of the passkey in the junk text (default: %d)", params.i_pos),
-        [](gpt_params & params, int value) {
+        string_format("position of the passkey in the junk text (default: %d)", params.i_pos),
+        [](common_params & params, int value) {
             params.i_pos = value;
         }
     ).set_examples({LLAMA_EXAMPLE_PASSKEY}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-o", "--output", "--output-file"}, "FNAME",
-        format("output file (default: '%s')",
+        string_format("output file (default: '%s')",
             ex == LLAMA_EXAMPLE_EXPORT_LORA
                 ? params.lora_outfile.c_str()
                 : ex == LLAMA_EXAMPLE_CVECTOR_GENERATOR
                     ? params.cvector_outfile.c_str()
                     : params.out_file.c_str()),
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             params.out_file = value;
             params.cvector_outfile = value;
             params.lora_outfile = value;
         }
     ).set_examples({LLAMA_EXAMPLE_IMATRIX, LLAMA_EXAMPLE_CVECTOR_GENERATOR, LLAMA_EXAMPLE_EXPORT_LORA}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-ofreq", "--output-frequency"}, "N",
-        format("output the imatrix every N iterations (default: %d)", params.n_out_freq),
-        [](gpt_params & params, int value) {
+        string_format("output the imatrix every N iterations (default: %d)", params.n_out_freq),
+        [](common_params & params, int value) {
             params.n_out_freq = value;
         }
     ).set_examples({LLAMA_EXAMPLE_IMATRIX}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--save-frequency"}, "N",
-        format("save an imatrix copy every N iterations (default: %d)", params.n_save_freq),
-        [](gpt_params & params, int value) {
+        string_format("save an imatrix copy every N iterations (default: %d)", params.n_save_freq),
+        [](common_params & params, int value) {
             params.n_save_freq = value;
         }
     ).set_examples({LLAMA_EXAMPLE_IMATRIX}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--process-output"},
-        format("collect data for the output tensor (default: %s)", params.process_output ? "true" : "false"),
-        [](gpt_params & params) {
+        string_format("collect data for the output tensor (default: %s)", params.process_output ? "true" : "false"),
+        [](common_params & params) {
             params.process_output = true;
         }
     ).set_examples({LLAMA_EXAMPLE_IMATRIX}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--no-ppl"},
-        format("do not compute perplexity (default: %s)", params.compute_ppl ? "true" : "false"),
-        [](gpt_params & params) {
+        string_format("do not compute perplexity (default: %s)", params.compute_ppl ? "true" : "false"),
+        [](common_params & params) {
             params.compute_ppl = false;
         }
     ).set_examples({LLAMA_EXAMPLE_IMATRIX}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--chunk", "--from-chunk"}, "N",
-        format("start processing the input from chunk N (default: %d)", params.i_chunk),
-        [](gpt_params & params, int value) {
+        string_format("start processing the input from chunk N (default: %d)", params.i_chunk),
+        [](common_params & params, int value) {
             params.i_chunk = value;
         }
     ).set_examples({LLAMA_EXAMPLE_IMATRIX}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-pps"},
-        format("is the prompt shared across parallel sequences (default: %s)", params.is_pp_shared ? "true" : "false"),
-        [](gpt_params & params) {
+        string_format("is the prompt shared across parallel sequences (default: %s)", params.is_pp_shared ? "true" : "false"),
+        [](common_params & params) {
             params.is_pp_shared = true;
         }
     ).set_examples({LLAMA_EXAMPLE_BENCH}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-npp"}, "n0,n1,...",
         "number of prompt tokens",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             auto p = string_split<int>(value, ',');
             params.n_pp.insert(params.n_pp.end(), p.begin(), p.end());
         }
     ).set_examples({LLAMA_EXAMPLE_BENCH}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-ntg"}, "n0,n1,...",
         "number of text generation tokens",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             auto p = string_split<int>(value, ',');
             params.n_tg.insert(params.n_tg.end(), p.begin(), p.end());
         }
     ).set_examples({LLAMA_EXAMPLE_BENCH}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-npl"}, "n0,n1,...",
         "number of parallel prompts",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             auto p = string_split<int>(value, ',');
             params.n_pl.insert(params.n_pl.end(), p.begin(), p.end());
         }
     ).set_examples({LLAMA_EXAMPLE_BENCH}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--embd-normalize"}, "N",
-        format("normalisation for embendings (default: %d) (-1=none, 0=max absolute int16, 1=taxicab, 2=euclidean, >2=p-norm)", params.embd_normalize),
-        [](gpt_params & params, int value) {
+        string_format("normalisation for embeddings (default: %d) (-1=none, 0=max absolute int16, 1=taxicab, 2=euclidean, >2=p-norm)", params.embd_normalize),
+        [](common_params & params, int value) {
             params.embd_normalize = value;
         }
     ).set_examples({LLAMA_EXAMPLE_EMBEDDING}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--embd-output-format"}, "FORMAT",
         "empty = default, \"array\" = [[],[]...], \"json\" = openai style, \"json+\" = same \"json\" + cosine similarity matrix",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             params.embd_out = value;
         }
     ).set_examples({LLAMA_EXAMPLE_EMBEDDING}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--embd-separator"}, "STRING",
-        "separator of embendings (default \\n) for example \"<#sep#>\"",
-        [](gpt_params & params, const std::string & value) {
+        "separator of embeddings (default \\n) for example \"<#sep#>\"",
+        [](common_params & params, const std::string & value) {
             params.embd_sep = value;
         }
     ).set_examples({LLAMA_EXAMPLE_EMBEDDING}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--host"}, "HOST",
-        format("ip address to listen (default: %s)", params.hostname.c_str()),
-        [](gpt_params & params, const std::string & value) {
+        string_format("ip address to listen (default: %s)", params.hostname.c_str()),
+        [](common_params & params, const std::string & value) {
             params.hostname = value;
         }
     ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_HOST"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--port"}, "PORT",
-        format("port to listen (default: %d)", params.port),
-        [](gpt_params & params, int value) {
+        string_format("port to listen (default: %d)", params.port),
+        [](common_params & params, int value) {
             params.port = value;
         }
     ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_PORT"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--path"}, "PATH",
-        format("path to serve static files from (default: %s)", params.public_path.c_str()),
-        [](gpt_params & params, const std::string & value) {
+        string_format("path to serve static files from (default: %s)", params.public_path.c_str()),
+        [](common_params & params, const std::string & value) {
             params.public_path = value;
         }
     ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_STATIC_PATH"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--embedding", "--embeddings"},
-        format("restrict to only support embedding use case; use only with dedicated embedding models (default: %s)", params.embedding ? "enabled" : "disabled"),
-        [](gpt_params & params) {
+        string_format("restrict to only support embedding use case; use only with dedicated embedding models (default: %s)", params.embedding ? "enabled" : "disabled"),
+        [](common_params & params) {
             params.embedding = true;
         }
     ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_EMBEDDINGS"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--reranking", "--rerank"},
-        format("enable reranking endpoint on server (default: %s)", params.reranking ? "enabled" : "disabled"),
-        [](gpt_params & params) {
+        string_format("enable reranking endpoint on server (default: %s)", params.reranking ? "enabled" : "disabled"),
+        [](common_params & params) {
             params.reranking = true;
         }
     ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_RERANKING"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--api-key"}, "KEY",
         "API key to use for authentication (default: none)",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             params.api_keys.push_back(value);
         }
     ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_API_KEY"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--api-key-file"}, "FNAME",
         "path to file containing API keys (default: none)",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             std::ifstream key_file(value);
             if (!key_file) {
-                throw std::runtime_error(format("error: failed to open file '%s'\n", value.c_str()));
+                throw std::runtime_error(string_format("error: failed to open file '%s'\n", value.c_str()));
             }
             std::string key;
             while (std::getline(key_file, key)) {
@@ -1785,70 +1827,74 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
             key_file.close();
         }
     ).set_examples({LLAMA_EXAMPLE_SERVER}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--ssl-key-file"}, "FNAME",
         "path to file a PEM-encoded SSL private key",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             params.ssl_file_key = value;
         }
     ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_SSL_KEY_FILE"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--ssl-cert-file"}, "FNAME",
         "path to file a PEM-encoded SSL certificate",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             params.ssl_file_cert = value;
         }
     ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_SSL_CERT_FILE"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-to", "--timeout"}, "N",
-        format("server read/write timeout in seconds (default: %d)", params.timeout_read),
-        [](gpt_params & params, int value) {
+        string_format("server read/write timeout in seconds (default: %d)", params.timeout_read),
+        [](common_params & params, int value) {
             params.timeout_read  = value;
             params.timeout_write = value;
         }
     ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_TIMEOUT"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--threads-http"}, "N",
-        format("number of threads used to process HTTP requests (default: %d)", params.n_threads_http),
-        [](gpt_params & params, int value) {
+        string_format("number of threads used to process HTTP requests (default: %d)", params.n_threads_http),
+        [](common_params & params, int value) {
             params.n_threads_http = value;
         }
     ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_THREADS_HTTP"));
-    add_opt(llama_arg(
-        {"-spf", "--system-prompt-file"}, "FNAME",
-        "set a file to load a system prompt (initial prompt of all slots), this is useful for chat applications",
-        [](gpt_params & params, const std::string & value) {
-            std::ifstream file(value);
-            if (!file) {
-                throw std::runtime_error(format("error: failed to open file '%s'\n", value.c_str()));
-            }
-            std::string system_prompt;
-            std::copy(
-                        std::istreambuf_iterator<char>(file),
-                        std::istreambuf_iterator<char>(),
-                        std::back_inserter(system_prompt)
-                        );
-            params.system_prompt = system_prompt;
-        }
-    ).set_examples({LLAMA_EXAMPLE_SERVER}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
+        {"--cache-reuse"}, "N",
+        string_format("min chunk size to attempt reusing from the cache via KV shifting (default: %d)", params.n_cache_reuse),
+        [](common_params & params, int value) {
+            params.n_cache_reuse = value;
+        }
+    ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_CACHE_REUSE"));
+    add_opt(common_arg(
         {"--metrics"},
-        format("enable prometheus compatible metrics endpoint (default: %s)", params.endpoint_metrics ? "enabled" : "disabled"),
-        [](gpt_params & params) {
+        string_format("enable prometheus compatible metrics endpoint (default: %s)", params.endpoint_metrics ? "enabled" : "disabled"),
+        [](common_params & params) {
             params.endpoint_metrics = true;
         }
     ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_ENDPOINT_METRICS"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
+        {"--slots"},
+        string_format("enable slots monitoring endpoint (default: %s)", params.endpoint_slots ? "enabled" : "disabled"),
+        [](common_params & params) {
+            params.endpoint_slots = true;
+        }
+    ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_ENDPOINT_SLOTS"));
+    add_opt(common_arg(
+        {"--props"},
+        string_format("enable changing global properties via POST /props (default: %s)", params.endpoint_props ? "enabled" : "disabled"),
+        [](common_params & params) {
+            params.endpoint_props = true;
+        }
+    ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_ENDPOINT_PROPS"));
+    add_opt(common_arg(
         {"--no-slots"},
-        format("disables slots monitoring endpoint (default: %s)", params.endpoint_slots ? "enabled" : "disabled"),
-        [](gpt_params & params) {
+        "disables slots monitoring endpoint",
+        [](common_params & params) {
             params.endpoint_slots = false;
         }
     ).set_examples({LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_NO_ENDPOINT_SLOTS"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--slot-save-path"}, "PATH",
         "path to save slot kv cache (default: disabled)",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             params.slot_save_path = value;
             // if doesn't end with DIRECTORY_SEPARATOR, add it
             if (!params.slot_save_path.empty() && params.slot_save_path[params.slot_save_path.size() - 1] != DIRECTORY_SEPARATOR) {
@@ -1856,14 +1902,14 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
             }
         }
     ).set_examples({LLAMA_EXAMPLE_SERVER}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--chat-template"}, "JINJA_TEMPLATE",
         "set custom jinja chat template (default: template taken from model's metadata)\n"
         "if suffix/prefix are specified, template will be disabled\n"
         "only commonly used templates are accepted:\nhttps://github.com/ggerganov/llama.cpp/wiki/Templates-supported-by-llama_chat_apply_template",
-        [](gpt_params & params, const std::string & value) {
-            if (!llama_chat_verify_template(value)) {
-                throw std::runtime_error(format(
+        [](common_params & params, const std::string & value) {
+            if (!common_chat_verify_template(value)) {
+                throw std::runtime_error(string_format(
                     "error: the supplied chat template is not supported: %s\n"
                     "note: llama.cpp does not use jinja parser, we only support commonly used templates\n",
                     value.c_str()
@@ -1872,133 +1918,122 @@ gpt_params_context gpt_params_parser_init(gpt_params & params, llama_example ex,
             params.chat_template = value;
         }
     ).set_examples({LLAMA_EXAMPLE_MAIN, LLAMA_EXAMPLE_SERVER}).set_env("LLAMA_ARG_CHAT_TEMPLATE"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-sps", "--slot-prompt-similarity"}, "SIMILARITY",
-        format("how much the prompt of a request must match the prompt of a slot in order to use that slot (default: %.2f, 0.0 = disabled)\n", params.slot_prompt_similarity),
-        [](gpt_params & params, const std::string & value) {
+        string_format("how much the prompt of a request must match the prompt of a slot in order to use that slot (default: %.2f, 0.0 = disabled)\n", params.slot_prompt_similarity),
+        [](common_params & params, const std::string & value) {
             params.slot_prompt_similarity = std::stof(value);
         }
     ).set_examples({LLAMA_EXAMPLE_SERVER}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--lora-init-without-apply"},
-        format("load LoRA adapters without applying them (apply later via POST /lora-adapters) (default: %s)", params.lora_init_without_apply ? "enabled" : "disabled"),
-        [](gpt_params & params) {
+        string_format("load LoRA adapters without applying them (apply later via POST /lora-adapters) (default: %s)", params.lora_init_without_apply ? "enabled" : "disabled"),
+        [](common_params & params) {
             params.lora_init_without_apply = true;
         }
     ).set_examples({LLAMA_EXAMPLE_SERVER}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--simple-io"},
         "use basic IO for better compatibility in subprocesses and limited consoles",
-        [](gpt_params & params) {
+        [](common_params & params) {
             params.simple_io = true;
         }
     ).set_examples({LLAMA_EXAMPLE_MAIN, LLAMA_EXAMPLE_INFILL}));
-    add_opt(llama_arg(
-        {"-ld", "--logdir"}, "LOGDIR",
-        "path under which to save YAML logs (no logging if unset)",
-        [](gpt_params & params, const std::string & value) {
-            params.logdir = value;
-            if (params.logdir.back() != DIRECTORY_SEPARATOR) {
-                params.logdir += DIRECTORY_SEPARATOR;
-            }
-        }
-    ));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--positive-file"}, "FNAME",
-        format("positive prompts file, one prompt per line (default: '%s')", params.cvector_positive_file.c_str()),
-        [](gpt_params & params, const std::string & value) {
+        string_format("positive prompts file, one prompt per line (default: '%s')", params.cvector_positive_file.c_str()),
+        [](common_params & params, const std::string & value) {
             params.cvector_positive_file = value;
         }
     ).set_examples({LLAMA_EXAMPLE_CVECTOR_GENERATOR}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--negative-file"}, "FNAME",
-        format("negative prompts file, one prompt per line (default: '%s')", params.cvector_negative_file.c_str()),
-        [](gpt_params & params, const std::string & value) {
+        string_format("negative prompts file, one prompt per line (default: '%s')", params.cvector_negative_file.c_str()),
+        [](common_params & params, const std::string & value) {
             params.cvector_negative_file = value;
         }
     ).set_examples({LLAMA_EXAMPLE_CVECTOR_GENERATOR}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--pca-batch"}, "N",
-        format("batch size used for PCA. Larger batch runs faster, but uses more memory (default: %d)", params.n_pca_batch),
-        [](gpt_params & params, int value) {
+        string_format("batch size used for PCA. Larger batch runs faster, but uses more memory (default: %d)", params.n_pca_batch),
+        [](common_params & params, int value) {
             params.n_pca_batch = value;
         }
     ).set_examples({LLAMA_EXAMPLE_CVECTOR_GENERATOR}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--pca-iter"}, "N",
-        format("number of iterations used for PCA (default: %d)", params.n_pca_iterations),
-        [](gpt_params & params, int value) {
+        string_format("number of iterations used for PCA (default: %d)", params.n_pca_iterations),
+        [](common_params & params, int value) {
             params.n_pca_iterations = value;
         }
     ).set_examples({LLAMA_EXAMPLE_CVECTOR_GENERATOR}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--method"}, "{pca, mean}",
         "dimensionality reduction method to be used (default: pca)",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             /**/ if (value == "pca") { params.cvector_dimre_method = DIMRE_METHOD_PCA; }
             else if (value == "mean") { params.cvector_dimre_method = DIMRE_METHOD_MEAN; }
             else { throw std::invalid_argument("invalid value"); }
         }
     ).set_examples({LLAMA_EXAMPLE_CVECTOR_GENERATOR}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--output-format"}, "{md,jsonl}",
         "output format for batched-bench results (default: md)",
-        [](gpt_params & params, const std::string & value) {
+        [](common_params & params, const std::string & value) {
             /**/ if (value == "jsonl") { params.batched_bench_output_jsonl = true; }
             else if (value == "md") { params.batched_bench_output_jsonl = false; }
             else { std::invalid_argument("invalid value"); }
         }
     ).set_examples({LLAMA_EXAMPLE_BENCH}));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--log-disable"},
         "Log disable",
-        [](gpt_params &) {
-            gpt_log_pause(gpt_log_main());
+        [](common_params &) {
+            common_log_pause(common_log_main());
         }
     ));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--log-file"}, "FNAME",
         "Log to file",
-        [](gpt_params &, const std::string & value) {
-            gpt_log_set_file(gpt_log_main(), value.c_str());
+        [](common_params &, const std::string & value) {
+            common_log_set_file(common_log_main(), value.c_str());
         }
     ));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--log-colors"},
         "Enable colored logging",
-        [](gpt_params &) {
-            gpt_log_set_colors(gpt_log_main(), true);
+        [](common_params &) {
+            common_log_set_colors(common_log_main(), true);
         }
     ).set_env("LLAMA_LOG_COLORS"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-v", "--verbose", "--log-verbose"},
         "Set verbosity level to infinity (i.e. log all messages, useful for debugging)",
-        [](gpt_params & params) {
+        [](common_params & params) {
             params.verbosity = INT_MAX;
-            gpt_log_set_verbosity_thold(INT_MAX);
+            common_log_set_verbosity_thold(INT_MAX);
         }
     ));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"-lv", "--verbosity", "--log-verbosity"}, "N",
         "Set the verbosity threshold. Messages with a higher verbosity will be ignored.",
-        [](gpt_params & params, int value) {
+        [](common_params & params, int value) {
             params.verbosity = value;
-            gpt_log_set_verbosity_thold(value);
+            common_log_set_verbosity_thold(value);
         }
     ).set_env("LLAMA_LOG_VERBOSITY"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--log-prefix"},
         "Enable prefx in log messages",
-        [](gpt_params &) {
-            gpt_log_set_prefix(gpt_log_main(), true);
+        [](common_params &) {
+            common_log_set_prefix(common_log_main(), true);
         }
     ).set_env("LLAMA_LOG_PREFIX"));
-    add_opt(llama_arg(
+    add_opt(common_arg(
         {"--log-timestamps"},
         "Enable timestamps in log messages",
-        [](gpt_params &) {
-            gpt_log_set_timestamps(gpt_log_main(), true);
+        [](common_params &) {
+            common_log_set_timestamps(common_log_main(), true);
         }
     ).set_env("LLAMA_LOG_TIMESTAMPS"));