cui-llama.rn 1.2.3 → 1.2.6

package/cpp/llama.h

@@ -2,6 +2,7 @@
 #define LLAMA_H
 
 #include "ggml.h"
+#include "ggml-cpu.h"
 #include "ggml-backend.h"
 
 #include <stddef.h>
@@ -206,7 +207,7 @@ extern "C" {
     enum llama_split_mode {
         LLAMA_SPLIT_MODE_NONE  = 0, // single GPU
         LLAMA_SPLIT_MODE_LAYER = 1, // split layers and KV across GPUs
-        LLAMA_SPLIT_MODE_ROW   = 2, // split rows across GPUs
+        LLAMA_SPLIT_MODE_ROW   = 2, // split layers and KV across GPUs, use tensor parallelism if supported
     };
 
     // TODO: simplify (https://github.com/ggerganov/llama.cpp/pull/9294#pullrequestreview-2286561979)
@@ -218,6 +219,7 @@ extern "C" {
 
     typedef struct llama_token_data_array {
         // TODO: consider SoA
+        // NOTE: this pointer can be modified by the samplers
         llama_token_data * data;
         size_t size;
         int64_t selected; // this is the index in the data array (i.e. not the token id)
@@ -233,8 +235,11 @@ extern "C" {
     // - token  : the token ids of the input (used when embd is NULL)
     // - embd   : token embeddings (i.e. float vector of size n_embd) (used when token is NULL)
     // - pos    : the positions of the respective token in the sequence
+    //            (if set to NULL, the token position will be tracked automatically by llama_decode)
     // - seq_id : the sequence to which the respective token belongs
+    //            (if set to NULL, the sequence ID will be assumed to be 0)
     // - logits : if zero, the logits (and/or the embeddings) for the respective token will not be output
+    //            (if set to NULL, only the logits for last token will be returned)
     //
     typedef struct llama_batch {
         int32_t n_tokens;
@@ -245,15 +250,6 @@ extern "C" {
         int32_t      *  n_seq_id;
         llama_seq_id ** seq_id;
         int8_t       *  logits; // TODO: rename this to "output"
-
-        // NOTE: helpers for smooth API transition - can be deprecated in the future
-        //       for future-proof code, use the above fields instead and ignore everything below
-        //
-        // pos[i] = all_pos_0 + i*all_pos_1
-        //
-        llama_pos    all_pos_0;  // used if pos == NULL
-        llama_pos    all_pos_1;  // used if pos == NULL
-        llama_seq_id all_seq_id; // used if seq_id == NULL
     } llama_batch;
 
     enum llama_model_kv_override_type {
@@ -280,10 +276,7 @@ extern "C" {
         int32_t n_gpu_layers; // number of layers to store in VRAM
         enum llama_split_mode split_mode; // how to split the model across multiple GPUs
 
-        // main_gpu interpretation depends on split_mode:
-        // LLAMA_SPLIT_MODE_NONE: the GPU that is used for the entire model
-        // LLAMA_SPLIT_MODE_ROW: the GPU that is used for small tensors and intermediate results
-        // LLAMA_SPLIT_MODE_LAYER: ignored
+        // the GPU that is used for the entire model when split_mode is LLAMA_SPLIT_MODE_NONE
         int32_t main_gpu;
 
         // proportion of the model (layers or rows) to offload to each GPU, size: llama_max_devices()
@@ -434,6 +427,7 @@ extern "C" {
     LLAMA_API bool llama_supports_mmap       (void);
     LLAMA_API bool llama_supports_mlock      (void);
     LLAMA_API bool llama_supports_gpu_offload(void);
+    LLAMA_API bool llama_supports_rpc        (void);
 
     LLAMA_API uint32_t llama_n_ctx      (const struct llama_context * ctx);
     LLAMA_API uint32_t llama_n_batch    (const struct llama_context * ctx);
@@ -776,15 +770,15 @@ extern "C" {
     // Decoding
     //
 
-    // Return batch for single sequence of tokens starting at pos_0
+    // Return batch for single sequence of tokens
+    // The sequence ID will be fixed to 0
+    // The position of the tokens will be tracked automatically by llama_decode
     //
     // NOTE: this is a helper function to facilitate transition to the new batch API - avoid using it
     //
     LLAMA_API struct llama_batch llama_batch_get_one(
                   llama_token * tokens,
-                      int32_t   n_tokens,
-                    llama_pos   pos_0,
-                 llama_seq_id   seq_id);
+                      int32_t   n_tokens);
 
     // Allocates a batch of tokens on the heap that can hold a maximum of n_tokens
     // Each token can be assigned up to n_seq_max sequence ids
@@ -897,6 +891,7 @@ extern "C" {
     // Special tokens
     LLAMA_API llama_token llama_token_bos(const struct llama_model * model); // beginning-of-sentence
     LLAMA_API llama_token llama_token_eos(const struct llama_model * model); // end-of-sentence
+    LLAMA_API llama_token llama_token_eot(const struct llama_model * model); // end-of-turn
     LLAMA_API llama_token llama_token_cls(const struct llama_model * model); // classification
     LLAMA_API llama_token llama_token_sep(const struct llama_model * model); // sentence separator
     LLAMA_API llama_token llama_token_nl (const struct llama_model * model); // next-line
@@ -905,11 +900,17 @@ extern "C" {
     LLAMA_API bool llama_add_bos_token(const struct llama_model * model);
     LLAMA_API bool llama_add_eos_token(const struct llama_model * model);
 
-    // Codellama infill tokens
-    LLAMA_API llama_token llama_token_prefix(const struct llama_model * model); // Beginning of infill prefix
-    LLAMA_API llama_token llama_token_middle(const struct llama_model * model); // Beginning of infill middle
-    LLAMA_API llama_token llama_token_suffix(const struct llama_model * model); // Beginning of infill suffix
-    LLAMA_API llama_token llama_token_eot   (const struct llama_model * model); // End of infill middle
+    // infill tokens
+    DEPRECATED(LLAMA_API llama_token llama_token_prefix(const struct llama_model * model), "use llama_token_fim_pre instead");
+    DEPRECATED(LLAMA_API llama_token llama_token_middle(const struct llama_model * model), "use llama_token_fim_mid instead");
+    DEPRECATED(LLAMA_API llama_token llama_token_suffix(const struct llama_model * model), "use llama_token_fim_suf instead");
+
+    LLAMA_API llama_token llama_token_fim_pre(const struct llama_model * model);
+    LLAMA_API llama_token llama_token_fim_suf(const struct llama_model * model);
+    LLAMA_API llama_token llama_token_fim_mid(const struct llama_model * model);
+    LLAMA_API llama_token llama_token_fim_pad(const struct llama_model * model);
+    LLAMA_API llama_token llama_token_fim_rep(const struct llama_model * model);
+    LLAMA_API llama_token llama_token_fim_sep(const struct llama_model * model);
 
     //
     // Tokenization
@@ -1068,12 +1069,13 @@ extern "C" {
 
     // available samplers:
 
-    LLAMA_API struct llama_sampler * llama_sampler_init_greedy     (void);
-    LLAMA_API struct llama_sampler * llama_sampler_init_dist       (uint32_t seed);
+    LLAMA_API struct llama_sampler * llama_sampler_init_greedy(void);
+    LLAMA_API struct llama_sampler * llama_sampler_init_dist  (uint32_t seed);
 
     /// @details Sorts candidate tokens by their logits in descending order and calculate probabilities based on logits.
     /// NOTE: Avoid using on the full vocabulary as the sorting can become slow. For example, apply top-k or top-p sampling first.
-    LLAMA_API struct llama_sampler * llama_sampler_init_softmax    (void);
+    DEPRECATED(LLAMA_API struct llama_sampler * llama_sampler_init_softmax    (void),
+        "will be removed in the future (see https://github.com/ggerganov/llama.cpp/pull/9896#discussion_r1800920915)");
 
     /// @details Top-K sampling described in academic paper "The Curious Case of Neural Text Degeneration" https://arxiv.org/abs/1904.09751
     LLAMA_API struct llama_sampler * llama_sampler_init_top_k      (int32_t k);
@@ -1084,19 +1086,18 @@ extern "C" {
     /// @details Minimum P sampling as described in https://github.com/ggerganov/llama.cpp/pull/3841
     LLAMA_API struct llama_sampler * llama_sampler_init_min_p      (float   p, size_t min_keep);
 
-    /// @details XTC sampling as described in https://github.com/oobabooga/text-generation-webui/pull/6335
-    LLAMA_API struct llama_sampler * llama_sampler_init_xtc        (float xtc_p, float xtc_t, size_t min_keep, uint32_t seed);
-
-    /// @details Tail Free Sampling described in https://www.trentonbricken.com/Tail-Free-Sampling/.
-    LLAMA_API struct llama_sampler * llama_sampler_init_tail_free  (float   z, size_t min_keep);
-
     /// @details Locally Typical Sampling implementation described in the paper https://arxiv.org/abs/2202.00666.
     LLAMA_API struct llama_sampler * llama_sampler_init_typical    (float   p, size_t min_keep);
+
+    /// #details Updates the logits l_i` = l_i/t. When t <= 0.0f, the maximum logit is kept at it's original value, the rest are set to -inf
     LLAMA_API struct llama_sampler * llama_sampler_init_temp       (float   t);
 
     /// @details Dynamic temperature implementation (a.k.a. entropy) described in the paper https://arxiv.org/abs/2309.02772.
     LLAMA_API struct llama_sampler * llama_sampler_init_temp_ext   (float   t, float   delta, float exponent);
 
+    /// @details XTC sampler as described in https://github.com/oobabooga/text-generation-webui/pull/6335
+    LLAMA_API struct llama_sampler * llama_sampler_init_xtc        (float   p, float   t,     size_t min_keep, uint32_t seed);
+
     /// @details Mirostat 1.0 algorithm described in the paper https://arxiv.org/abs/2007.14966. Uses tokens instead of words.
     /// @param candidates A vector of `llama_token_data` containing the candidate tokens, their probabilities (p), and log-odds (logit) for the current position in the generated text.
     /// @param tau  The target cross-entropy (or surprise) value you want to achieve for the generated text. A higher value corresponds to more surprising or less predictable text, while a lower value corresponds to less surprising or more predictable text.
@@ -1136,11 +1137,43 @@ extern "C" {
                                 bool   penalize_nl,     // consider newlines as a repeatable token
                                 bool   ignore_eos);     // ignore the end-of-sequence token
 
+    ///  @details DRY sampler, designed by p-e-w, as described in: https://github.com/oobabooga/text-generation-webui/pull/5677, porting Koboldcpp implementation authored by pi6am: https://github.com/LostRuins/koboldcpp/pull/982
+    LLAMA_API struct llama_sampler *    llama_sampler_init_dry(
+            const struct llama_model *  model,
+                               float    dry_multiplier,
+                               float    dry_base,
+                             int32_t    dry_allowed_length,
+                             int32_t    dry_penalty_last_n,
+                          const char ** seq_breakers,
+                              size_t    num_breakers);
+
     LLAMA_API struct llama_sampler * llama_sampler_init_logit_bias(
                              int32_t   n_vocab,
                              int32_t   n_logit_bias,
               const llama_logit_bias * logit_bias);
 
+    // this sampler is meant to be used for fill-in-the-middle infilling
+    // it's supposed to be used after top_k + top_p sampling
+    //
+    // 1. if the sum of the EOG probs times the number of candidates is higher than the sum of the other probs -> pick EOG
+    // 2. combine probs of tokens that have the same prefix
+    //
+    // example:
+    //
+    // - before:
+    //   "hel":   0.5
+    //   "hell":  0.2
+    //   "hello": 0.1
+    //   "dummy": 0.1
+    //
+    // - after:
+    //   "hel":   0.8
+    //   "dummy": 0.1
+    //
+    // 3. discard non-EOG tokens with low prob
+    // 4. if no tokens are left -> pick EOT
+    //
+    LLAMA_API struct llama_sampler * llama_sampler_init_infill(const struct llama_model * model);
 
     // Returns the seed used by the sampler if applicable, LLAMA_DEFAULT_SEED otherwise
     LLAMA_API uint32_t llama_sampler_get_seed(const struct llama_sampler * smpl);

package/cpp/log.cpp

@@ -8,10 +8,10 @@
 #include <thread>
 #include <vector>
 
-int gpt_log_verbosity_thold = LOG_DEFAULT_LLAMA;
+int common_log_verbosity_thold = LOG_DEFAULT_LLAMA;
 
-void gpt_log_set_verbosity_thold(int verbosity) {
-    gpt_log_verbosity_thold = verbosity;
+void common_log_set_verbosity_thold(int verbosity) {
+    common_log_verbosity_thold = verbosity;
 }
 
 #define LOG_COL_DEFAULT "\033[0m"
@@ -29,16 +29,16 @@ static int64_t t_us() {
 }
 
 // colors
-enum gpt_log_col : int {
-    GPT_LOG_COL_DEFAULT = 0,
-    GPT_LOG_COL_BOLD,
-    GPT_LOG_COL_RED,
-    GPT_LOG_COL_GREEN,
-    GPT_LOG_COL_YELLOW,
-    GPT_LOG_COL_BLUE,
-    GPT_LOG_COL_MAGENTA,
-    GPT_LOG_COL_CYAN,
-    GPT_LOG_COL_WHITE,
+enum common_log_col : int {
+    COMMON_LOG_COL_DEFAULT = 0,
+    COMMON_LOG_COL_BOLD,
+    COMMON_LOG_COL_RED,
+    COMMON_LOG_COL_GREEN,
+    COMMON_LOG_COL_YELLOW,
+    COMMON_LOG_COL_BLUE,
+    COMMON_LOG_COL_MAGENTA,
+    COMMON_LOG_COL_CYAN,
+    COMMON_LOG_COL_WHITE,
 };
 
 // disable colors by default
@@ -54,7 +54,7 @@ static std::vector<const char *> g_col = {
     "",
 };
 
-struct gpt_log_entry {
+struct common_log_entry {
     enum lm_ggml_log_level level;
 
     bool prefix;
@@ -71,7 +71,7 @@ struct gpt_log_entry {
         if (!fcur) {
             // stderr displays DBG messages only when their verbosity level is not higher than the threshold
             // these messages will still be logged to a file
-            if (level == LM_GGML_LOG_LEVEL_DEBUG && gpt_log_verbosity_thold < LOG_DEFAULT_DEBUG) {
+            if (level == LM_GGML_LOG_LEVEL_DEBUG && common_log_verbosity_thold < LOG_DEFAULT_DEBUG) {
                 return;
             }
 
@@ -86,19 +86,19 @@ struct gpt_log_entry {
             if (timestamp) {
                 // [M.s.ms.us]
                 fprintf(fcur, "%s%d.%02d.%03d.%03d%s ",
-                        g_col[GPT_LOG_COL_BLUE],
+                        g_col[COMMON_LOG_COL_BLUE],
                         (int) (timestamp / 1000000 / 60),
                         (int) (timestamp / 1000000 % 60),
                         (int) (timestamp / 1000 % 1000),
                         (int) (timestamp % 1000),
-                        g_col[GPT_LOG_COL_DEFAULT]);
+                        g_col[COMMON_LOG_COL_DEFAULT]);
             }
 
             switch (level) {
-                case LM_GGML_LOG_LEVEL_INFO:  fprintf(fcur, "%sI %s", g_col[GPT_LOG_COL_GREEN],   g_col[GPT_LOG_COL_DEFAULT]); break;
-                case LM_GGML_LOG_LEVEL_WARN:  fprintf(fcur, "%sW %s", g_col[GPT_LOG_COL_MAGENTA], ""                        ); break;
-                case LM_GGML_LOG_LEVEL_ERROR: fprintf(fcur, "%sE %s", g_col[GPT_LOG_COL_RED],     ""                        ); break;
-                case LM_GGML_LOG_LEVEL_DEBUG: fprintf(fcur, "%sD %s", g_col[GPT_LOG_COL_YELLOW],  ""                        ); break;
+                case LM_GGML_LOG_LEVEL_INFO:  fprintf(fcur, "%sI %s", g_col[COMMON_LOG_COL_GREEN],   g_col[COMMON_LOG_COL_DEFAULT]); break;
+                case LM_GGML_LOG_LEVEL_WARN:  fprintf(fcur, "%sW %s", g_col[COMMON_LOG_COL_MAGENTA], ""                        ); break;
+                case LM_GGML_LOG_LEVEL_ERROR: fprintf(fcur, "%sE %s", g_col[COMMON_LOG_COL_RED],     ""                        ); break;
+                case LM_GGML_LOG_LEVEL_DEBUG: fprintf(fcur, "%sD %s", g_col[COMMON_LOG_COL_YELLOW],  ""                        ); break;
                 default:
                     break;
             }
@@ -107,18 +107,18 @@ struct gpt_log_entry {
         fprintf(fcur, "%s", msg.data());
 
         if (level == LM_GGML_LOG_LEVEL_WARN || level == LM_GGML_LOG_LEVEL_ERROR || level == LM_GGML_LOG_LEVEL_DEBUG) {
-            fprintf(fcur, "%s", g_col[GPT_LOG_COL_DEFAULT]);
+            fprintf(fcur, "%s", g_col[COMMON_LOG_COL_DEFAULT]);
         }
 
         fflush(fcur);
     }
 };
 
-struct gpt_log {
+struct common_log {
     // default capacity - will be expanded if needed
-    gpt_log() : gpt_log(256) {}
+    common_log() : common_log(256) {}
 
-    gpt_log(size_t capacity) {
+    common_log(size_t capacity) {
         file = nullptr;
         prefix = false;
         timestamps = false;
@@ -137,7 +137,7 @@ struct gpt_log {
         resume();
     }
 
-    ~gpt_log() {
+    ~common_log() {
         pause();
         if (file) {
             fclose(file);
@@ -158,12 +158,12 @@ private:
     int64_t t_start;
 
     // ring buffer of entries
-    std::vector<gpt_log_entry> entries;
+    std::vector<common_log_entry> entries;
     size_t head;
     size_t tail;
 
     // worker thread copies into this
-    gpt_log_entry cur;
+    common_log_entry cur;
 
 public:
     void add(enum lm_ggml_log_level level, const char * fmt, va_list args) {
@@ -219,7 +219,7 @@ public:
         tail = (tail + 1) % entries.size();
         if (tail == head) {
             // expand the buffer
-            std::vector<gpt_log_entry> new_entries(2*entries.size());
+            std::vector<common_log_entry> new_entries(2*entries.size());
 
             size_t new_tail = 0;
 
@@ -320,15 +320,15 @@ public:
         pause();
 
         if (colors) {
-            g_col[GPT_LOG_COL_DEFAULT] = LOG_COL_DEFAULT;
-            g_col[GPT_LOG_COL_BOLD]    = LOG_COL_BOLD;
-            g_col[GPT_LOG_COL_RED]     = LOG_COL_RED;
-            g_col[GPT_LOG_COL_GREEN]   = LOG_COL_GREEN;
-            g_col[GPT_LOG_COL_YELLOW]  = LOG_COL_YELLOW;
-            g_col[GPT_LOG_COL_BLUE]    = LOG_COL_BLUE;
-            g_col[GPT_LOG_COL_MAGENTA] = LOG_COL_MAGENTA;
-            g_col[GPT_LOG_COL_CYAN]    = LOG_COL_CYAN;
-            g_col[GPT_LOG_COL_WHITE]   = LOG_COL_WHITE;
+            g_col[COMMON_LOG_COL_DEFAULT] = LOG_COL_DEFAULT;
+            g_col[COMMON_LOG_COL_BOLD]    = LOG_COL_BOLD;
+            g_col[COMMON_LOG_COL_RED]     = LOG_COL_RED;
+            g_col[COMMON_LOG_COL_GREEN]   = LOG_COL_GREEN;
+            g_col[COMMON_LOG_COL_YELLOW]  = LOG_COL_YELLOW;
+            g_col[COMMON_LOG_COL_BLUE]    = LOG_COL_BLUE;
+            g_col[COMMON_LOG_COL_MAGENTA] = LOG_COL_MAGENTA;
+            g_col[COMMON_LOG_COL_CYAN]    = LOG_COL_CYAN;
+            g_col[COMMON_LOG_COL_WHITE]   = LOG_COL_WHITE;
         } else {
             for (size_t i = 0; i < g_col.size(); i++) {
                 g_col[i] = "";
@@ -355,47 +355,47 @@ public:
 // public API
 //
 
-struct gpt_log * gpt_log_init() {
-    return new gpt_log;
+struct common_log * common_log_init() {
+    return new common_log;
 }
 
-struct gpt_log * gpt_log_main() {
-    static struct gpt_log log;
+struct common_log * common_log_main() {
+    static struct common_log log;
 
     return &log;
 }
 
-void gpt_log_pause(struct gpt_log * log) {
+void common_log_pause(struct common_log * log) {
     log->pause();
 }
 
-void gpt_log_resume(struct gpt_log * log) {
+void common_log_resume(struct common_log * log) {
     log->resume();
 }
 
-void gpt_log_free(struct gpt_log * log) {
+void common_log_free(struct common_log * log) {
     delete log;
 }
 
-void gpt_log_add(struct gpt_log * log, enum lm_ggml_log_level level, const char * fmt, ...) {
+void common_log_add(struct common_log * log, enum lm_ggml_log_level level, const char * fmt, ...) {
     va_list args;
     va_start(args, fmt);
     log->add(level, fmt, args);
     va_end(args);
 }
 
-void gpt_log_set_file(struct gpt_log * log, const char * file) {
+void common_log_set_file(struct common_log * log, const char * file) {
     log->set_file(file);
 }
 
-void gpt_log_set_colors(struct gpt_log * log, bool colors) {
+void common_log_set_colors(struct common_log * log, bool colors) {
     log->set_colors(colors);
 }
 
-void gpt_log_set_prefix(struct gpt_log * log, bool prefix) {
+void common_log_set_prefix(struct common_log * log, bool prefix) {
     log->set_prefix(prefix);
 }
 
-void gpt_log_set_timestamps(struct gpt_log * log, bool timestamps) {
+void common_log_set_timestamps(struct common_log * log, bool timestamps) {
     log->set_timestamps(timestamps);
 }

package/cpp/log.h

@@ -14,23 +14,23 @@
 #define LOG_DEFAULT_LLAMA 0
 
 // needed by the LOG_TMPL macro to avoid computing log arguments if the verbosity lower
-// set via gpt_log_set_verbosity()
-extern int gpt_log_verbosity_thold;
+// set via common_log_set_verbosity()
+extern int common_log_verbosity_thold;
 
-void gpt_log_set_verbosity_thold(int verbosity); // not thread-safe
+void common_log_set_verbosity_thold(int verbosity); // not thread-safe
 
-// the gpt_log uses an internal worker thread to print/write log messages
+// the common_log uses an internal worker thread to print/write log messages
 // when the worker thread is paused, incoming log messages are discarded
-struct gpt_log;
+struct common_log;
 
-struct gpt_log * gpt_log_init();
-struct gpt_log * gpt_log_main(); // singleton, automatically destroys itself on exit
-void             gpt_log_pause (struct gpt_log * log); // pause  the worker thread, not thread-safe
-void             gpt_log_resume(struct gpt_log * log); // resume the worker thread, not thread-safe
-void             gpt_log_free  (struct gpt_log * log);
+struct common_log * common_log_init();
+struct common_log * common_log_main(); // singleton, automatically destroys itself on exit
+void                common_log_pause (struct common_log * log); // pause  the worker thread, not thread-safe
+void                common_log_resume(struct common_log * log); // resume the worker thread, not thread-safe
+void                common_log_free  (struct common_log * log);
 
 LOG_ATTRIBUTE_FORMAT(3, 4)
-void gpt_log_add(struct gpt_log * log, enum lm_ggml_log_level level, const char * fmt, ...);
+void common_log_add(struct common_log * log, enum lm_ggml_log_level level, const char * fmt, ...);
 
 // defaults: file = NULL, colors = false, prefix = false, timestamps = false
 //
@@ -54,10 +54,10 @@ void gpt_log_add(struct gpt_log * log, enum lm_ggml_log_level level, const char
 // D - debug   (stderr, V = LOG_DEFAULT_DEBUG)
 //
 
-void gpt_log_set_file      (struct gpt_log * log, const char * file);       // not thread-safe
-void gpt_log_set_colors    (struct gpt_log * log,       bool   colors);     // not thread-safe
-void gpt_log_set_prefix    (struct gpt_log * log,       bool   prefix);     // whether to output prefix to each log
-void gpt_log_set_timestamps(struct gpt_log * log,       bool   timestamps); // whether to output timestamps in the prefix
+void common_log_set_file      (struct common_log * log, const char * file);       // not thread-safe
+void common_log_set_colors    (struct common_log * log,       bool   colors);     // not thread-safe
+void common_log_set_prefix    (struct common_log * log,       bool   prefix);     // whether to output prefix to each log
+void common_log_set_timestamps(struct common_log * log,       bool   timestamps); // whether to output timestamps in the prefix
 
 // helper macros for logging
 // use these to avoid computing log arguments if the verbosity of the log is higher than the threshold
@@ -66,7 +66,7 @@ void gpt_log_set_timestamps(struct gpt_log * log,       bool   timestamps); // w
 //
 //   LOG_DBG("this is a debug message: %d\n", expensive_function());
 //
-// this will avoid calling expensive_function() if LOG_DEFAULT_DEBUG > gpt_log_verbosity_thold
+// this will avoid calling expensive_function() if LOG_DEFAULT_DEBUG > common_log_verbosity_thold
 //
 
 
@@ -98,8 +98,8 @@ void gpt_log_set_timestamps(struct gpt_log * log,       bool   timestamps); // w
 
 #define LOG_TMPL(level, verbosity, ...) \
     do { \
-        if ((verbosity) <= gpt_log_verbosity_thold) { \
-            gpt_log_add(gpt_log_main(), (level), __VA_ARGS__); \
+        if ((verbosity) <= common_log_verbosity_thold) { \
+            common_log_add(common_log_main(), (level), __VA_ARGS__); \
         } \
     } while (0)
 

package/cpp/rn-llama.hpp

@@ -117,7 +117,7 @@ static size_t find_partial_stop_string(const std::string &stop,
 // format incomplete utf-8 multibyte character for output
 static std::string tokens_to_output_formatted_string(const llama_context *ctx, const llama_token token)
 {
-    std::string out = token == -1 ? "" : llama_token_to_piece(ctx, token);
+    std::string out = token == -1 ? "" : common_token_to_piece(ctx, token);
     // if the size is 1 and first bit is 1, meaning it's a partial character
     //   (size > 1 meaning it's already a known token)
     if (out.size() == 1 && (out[0] & 0x80) == 0x80)
@@ -136,7 +136,7 @@ static std::string tokens_to_str(llama_context *ctx, Iter begin, Iter end)
     std::string ret;
     for (; begin != end; ++begin)
     {
-        ret += llama_token_to_piece(ctx, *begin);
+        ret += common_token_to_piece(ctx, *begin);
     }
     return ret;
 }
@@ -157,11 +157,11 @@ struct llama_rn_context
 
     std::vector<llama_token> embd;
 
-    gpt_params params;
+    common_params params;
 
     llama_model *model = nullptr;
     llama_context *ctx = nullptr;
-    gpt_sampler *ctx_sampling = nullptr;
+    common_sampler *ctx_sampling = nullptr;
   
     int n_ctx;
 
@@ -186,7 +186,7 @@ struct llama_rn_context
         }
         if (ctx_sampling != nullptr)
         {
-            gpt_sampler_free(ctx_sampling);
+            common_sampler_free(ctx_sampling);
         }
     }
 
@@ -213,16 +213,16 @@ struct llama_rn_context
 
     bool initSampling() {
         if (ctx_sampling != nullptr) {
-            gpt_sampler_free(ctx_sampling);
+            common_sampler_free(ctx_sampling);
         }
-        ctx_sampling = gpt_sampler_init(model, params.sparams);
+        ctx_sampling = common_sampler_init(model, params.sparams);
         return ctx_sampling != nullptr;
     }
 
-    bool loadModel(gpt_params &params_)
+    bool loadModel(common_params &params_)
     {
         params = params_;
-        llama_init_result result = llama_init_from_gpt_params(params);
+        common_init_result result = common_init_from_params(params);
         model = result.model;
         ctx = result.context;
         if (model == nullptr)
@@ -268,7 +268,7 @@ struct llama_rn_context
 
     void loadPrompt()
     {
-        std::vector<llama_token> prompt_tokens = ::llama_tokenize(ctx, params.prompt, true, true);
+        std::vector<llama_token> prompt_tokens = ::common_tokenize(ctx, params.prompt, true, true);
         num_prompt_tokens = prompt_tokens.size();
 
         // LOG tokens
@@ -302,7 +302,7 @@ struct llama_rn_context
         // push the prompt into the sampling context (do not apply grammar)
         for (auto & token : prompt_tokens)
         {
-           gpt_sampler_accept(ctx_sampling, token, false);
+           common_sampler_accept(ctx_sampling, token, false);
         }
         // compare the evaluated prompt with the new prompt
         n_past = params.embedding? 0 :  common_part(embd, prompt_tokens);
@@ -375,8 +375,8 @@ struct llama_rn_context
             {
                 n_eval = params.n_batch;
             }
-            if (llama_decode(ctx, llama_batch_get_one(&embd[n_past], n_eval, n_past, 0)))
-            {
+            if (llama_decode(ctx, llama_batch_get_one(&embd[n_past], n_eval)))
+            {   
                 LOG_ERROR("failed to eval, n_eval: %d, n_past: %d, n_threads: %d, embd: %s",
                     n_eval,
                     n_past,
@@ -408,18 +408,19 @@ struct llama_rn_context
             std::vector<llama_token_data> candidates;
             candidates.reserve(llama_n_vocab(model));
 
-            result.tok = gpt_sampler_sample(ctx_sampling, ctx, -1);
+            result.tok = common_sampler_sample(ctx_sampling, ctx, -1);
             
-            llama_token_data_array cur_p = *gpt_sampler_get_candidates(ctx_sampling);
+            llama_token_data_array cur_p = *common_sampler_get_candidates(ctx_sampling);
 
             const int32_t n_probs = params.sparams.n_probs;
             
-            
-            if (params.sparams.temp <= 0 && n_probs > 0)
+            // deprecated
+            /*if (params.sparams.temp <= 0 && n_probs > 0)
             {
                 // For llama_sample_token_greedy we need to sort candidates
                 llama_sampler_init_softmax();
-            }
+
+            }*/
             
 
             for (size_t i = 0; i < std::min(cur_p.size, (size_t)n_probs); ++i)
@@ -427,7 +428,7 @@ struct llama_rn_context
                 result.probs.push_back({cur_p.data[i].id, cur_p.data[i].p});
             }
 
-            gpt_sampler_accept(ctx_sampling, result.tok, true);
+            common_sampler_accept(ctx_sampling, result.tok, true);
             if (tg) {
                 num_tokens_predicted++;
             }
@@ -487,7 +488,7 @@ struct llama_rn_context
     {
         const completion_token_output token_with_probs = nextToken();
 
-        const std::string token_text = token_with_probs.tok == -1 ? "" : llama_token_to_piece(ctx, token_with_probs.tok);
+        const std::string token_text = token_with_probs.tok == -1 ? "" : common_token_to_piece(ctx, token_with_probs.tok);
         generated_text += token_text;
 
         if (params.sparams.n_probs > 0)
@@ -528,7 +529,7 @@ struct llama_rn_context
         }
 
         LOG_VERBOSE("next token, token: %s, token_text: %s, has_next_token: %d, n_remain: %d, num_tokens_predicted: %d, stopped_eos: %d, stopped_word: %d, stopped_limit: %d, stopping_word: %s",
-            llama_token_to_piece(ctx, token_with_probs.tok),
+            common_token_to_piece(ctx, token_with_probs.tok),
             tokens_to_output_formatted_string(ctx, token_with_probs.tok).c_str(),
             has_next_token,
             n_remain,
@@ -562,7 +563,7 @@ struct llama_rn_context
             return std::vector<float>(n_embd, 0.0f);
         }
         std::vector<float> embedding(data, data + n_embd), out(data, data + n_embd);
-        llama_embd_normalize(embedding.data(), out.data(), n_embd, params.embd_normalize);
+        common_embd_normalize(embedding.data(), out.data(), n_embd, params.embd_normalize);
         return out;
     }