whispercpp 1.3.5 → 1.3.6

data/ext/sources/examples/talk-llama/llama-vocab.cpp

@@ -90,7 +90,7 @@ static_assert(std::is_trivially_copyable<llm_symbol>::value, "llm_symbol is not
 //
 // SPM tokenizer
 // original implementation:
-// https://github.com/ggerganov/llama.cpp/commit/074bea2eb1f1349a0118239c4152914aecaa1be4
+// https://github.com/ggml-org/llama.cpp/commit/074bea2eb1f1349a0118239c4152914aecaa1be4
 //
 
 struct llm_bigram_spm {
@@ -285,10 +285,19 @@ struct llm_tokenizer_bpe : llm_tokenizer {
                     // original regex from tokenizer.json
                     //"(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?\\p{L}+|\\p{N}{1,3}| ?[^\\s\\p{L}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
 
-                    // adapted: https://github.com/ggerganov/llama.cpp/pull/6920#issuecomment-2080233989
+                    // adapted: https://github.com/ggml-org/llama.cpp/pull/6920#issuecomment-2080233989
                     "(?:'[sS]|'[tT]|'[rR][eE]|'[vV][eE]|'[mM]|'[lL][lL]|'[dD])|[^\\r\\n\\p{L}\\p{N}]?\\p{L}+|\\p{N}{1,3}| ?[^\\s\\p{L}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
                 };
                 break;
+            case LLAMA_VOCAB_PRE_TYPE_JAIS2:
+                regex_exprs = {
+                    // original regex from tokenizer.json
+                    //"(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?\\p{L}+|\\p{N}{1,3}| ?[^\\s\\p{L}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s{512}(?!\\S)|\\s{256}(?!\\S)|\\s{128}(?!\\S)|\\s{64}(?!\\S)|\\s{32}(?!\\S)|\\s{16}(?!\\S)|\\s{8}(?!\\S)|\\s{4}(?!\\S)|\\s{1,2}(?!\\S)|\\s{1}",
+
+                    // adapted: same as llama3 but with cascading whitespace pattern
+                    "(?:'[sS]|'[tT]|'[rR][eE]|'[vV][eE]|'[mM]|'[lL][lL]|'[dD])|[^\\r\\n\\p{L}\\p{N}]?\\p{L}+|\\p{N}{1,3}| ?[^\\s\\p{L}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s{512}(?!\\S)|\\s{256}(?!\\S)|\\s{128}(?!\\S)|\\s{64}(?!\\S)|\\s{32}(?!\\S)|\\s{16}(?!\\S)|\\s{8}(?!\\S)|\\s{4}(?!\\S)|\\s{1,2}(?!\\S)|\\s{1}",
+                };
+                break;
             case LLAMA_VOCAB_PRE_TYPE_DBRX:
             case LLAMA_VOCAB_PRE_TYPE_SMAUG:
                 regex_exprs = {
@@ -308,6 +317,7 @@ struct llm_tokenizer_bpe : llm_tokenizer {
                 break;
             case LLAMA_VOCAB_PRE_TYPE_DEEPSEEK3_LLM:
             case LLAMA_VOCAB_PRE_TYPE_HUNYUAN_DENSE:
+            case LLAMA_VOCAB_PRE_TYPE_JOYAI_LLM:
                 regex_exprs = {
                     "\\p{N}{1,3}",
                     "[一-龥぀-ゟ゠-ヿ]+",
@@ -368,6 +378,13 @@ struct llm_tokenizer_bpe : llm_tokenizer {
                     "(?:'[sS]|'[tT]|'[rR][eE]|'[vV][eE]|'[mM]|'[lL][lL]|'[dD])|[^\\r\\n\\p{L}\\p{N}]?\\p{L}+|\\p{N}| ?[^\\s\\p{L}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
                 };
                 break;
+            case LLAMA_VOCAB_PRE_TYPE_QWEN35:
+                regex_exprs = {
+                    // original regex from tokenizer.json
+                    // "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+"
+                    "(?:'[sS]|'[tT]|'[rR][eE]|'[vV][eE]|'[mM]|'[lL][lL]|'[dD])|[^\\r\\n\\p{L}\\p{N}]?[\\p{L}\\p{M}]+|\\p{N}| ?[^\\s\\p{L}\\p{M}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+                };
+                break;
             case LLAMA_VOCAB_PRE_TYPE_PORO:
             case LLAMA_VOCAB_PRE_TYPE_BLOOM:
             case LLAMA_VOCAB_PRE_TYPE_GPT3_FINNISH:
@@ -415,6 +432,14 @@ struct llm_tokenizer_bpe : llm_tokenizer {
                     "[^\\r\\n\\p{L}\\p{N}]?((?=[\\p{L}])([^a-z]))*((?=[\\p{L}])([^A-Z]))+(?:'[sS]|'[tT]|'[rR][eE]|'[vV][eE]|'[mM]|'[lL][lL]|'[dD])?|[^\\r\\n\\p{L}\\p{N}]?((?=[\\p{L}])([^a-z]))+((?=[\\p{L}])([^A-Z]))*(?:'[sS]|'[tT]|'[rR][eE]|'[vV][eE]|'[mM]|'[lL][lL]|'[dD])?|\\p{N}{1,3}| ?[^\\s\\p{L}\\p{N}]+[\\r\\n/]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
                 };
                 break;
+            case LLAMA_VOCAB_PRE_TYPE_TINY_AYA:
+                regex_exprs = {
+                    // original regex from tokenizer.json: "\\d{1,3}(?=(?:\\d{3})*\\b)"
+                    "\\d{1,3}(?=(?:\\d{3})*\\b)",
+                    // original regex from tokenizer.json: "[^\\r\\n\\p{L}\\p{N}]?[\\p{Lu}\\p{Lt}\\p{Lm}\\p{Lo}\\p{M}]*[\\p{Ll}\\p{Lm}\\p{Lo}\\p{M}]+(?i:'s|'t|'re|'ve|'m|'ll|'d)?|[^\\r\\n\\p{L}\\p{N}]?[\\p{Lu}\\p{Lt}\\p{Lm}\\p{Lo}\\p{M}]+[\\p{Ll}\\p{Lm}\\p{Lo}\\p{M}]*(?i:'s|'t|'re|'ve|'m|'ll|'d)?|\\p{N}{1,3}| ?[^\\s\\p{L}\\p{N}]+[\\r\\n/]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+"
+                    "[^\\r\\n\\p{L}\\p{N}]?[\\p{Lu}\\p{Lt}\\p{Lm}\\p{Lo}\\p{M}]*[\\p{Ll}\\p{Lm}\\p{Lo}\\p{M}]+(?:'[sS]|'[tT]|'[rR][eE]|'[vV][eE]|'[mM]|'[lL][lL]|'[dD])?|[^\\r\\n\\p{L}\\p{N}]?[\\p{Lu}\\p{Lt}\\p{Lm}\\p{Lo}\\p{M}]+[\\p{Ll}\\p{Lm}\\p{Lo}\\p{M}]*(?:'[sS]|'[tT]|'[rR][eE]|'[vV][eE]|'[mM]|'[lL][lL]|'[dD])?|\\p{N}{1,3}| ?[^\\s\\p{L}\\p{N}]+[\\r\\n/]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+                };
+                break;
             case LLAMA_VOCAB_PRE_TYPE_KIMI_K2:
                 regex_exprs = {
                     // K2 trigger pattern - this will activate the custom K2 handler in unicode.cpp
@@ -461,6 +486,13 @@ struct llm_tokenizer_bpe : llm_tokenizer {
                     "[!\"#$%&'()*+,\\-./:;<=>?@\\[\\\\\\]^_`{|}~][A-Za-z]+|[^\\r\\n\\p{L}\\p{P}\\p{S}]?[\\p{L}\\p{M}]+| ?[\\p{P}\\p{S}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
                 };
                 break;
+            case LLAMA_VOCAB_PRE_TYPE_EXAONE_MOE:
+                regex_exprs = {
+                    // original regex from tokenizer.json
+                    // "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?(?:\\p{L}\\p{M}*(?: \\p{L}\\p{M}*)*)+|\\p{N}| ?[^\\s\\p{L}\\p{N}]+[\\r\\n/]?|\\s*[\\r\\n]|\\s+(?!\\S)|\\s+"
+                    "(?:'[sS]|'[tT]|'[rR][eE]|'[vV][eE]|'[mM]|'[lL][lL]|'[dD])|[^\\r\\n\\p{L}\\p{N}]?(?:\\p{L}\\p{M}*(?: \\p{L}\\p{M}*)*)+|\\p{N}| ?[^\\s\\p{L}\\p{N}]+[\\r\\n/]?|\\s*[\\r\\n]|\\s+(?!\\S)|\\s+",
+                };
+                break;
             default:
                 // default regex for BPE tokenization pre-processing
                 regex_exprs = {
@@ -1687,7 +1719,7 @@ private:
 };
 
 void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
-    struct gguf_context * ctx = ml.meta.get();
+    struct gguf_context * ctx = ml.metadata;
 
     // determine vocab type
     {
@@ -1745,26 +1777,33 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
 
             // read bpe merges and populate bpe ranks
             const int merges_keyidx = gguf_find_key(ctx, kv(LLM_KV_TOKENIZER_MERGES).c_str());
+            // Kimi-K2 uses custom tokenization without traditional BPE merges
+            const bool is_kimi_k2 = (tokenizer_pre == "kimi-k2");
+
             if (merges_keyidx == -1) {
-                throw std::runtime_error("cannot find tokenizer merges in model file\n");
-            }
+                if (!is_kimi_k2) {
+                    throw std::runtime_error("cannot find tokenizer merges in model file\n");
+                }
+                // Kimi-K2 doesn't need merges, skip
+                LLAMA_LOG_INFO("%s: Kimi-K2 tokenizer detected, skipping BPE merges\n", __func__);
+            } else {
+                const int n_merges = gguf_get_arr_n(ctx, merges_keyidx);
+                for (int i = 0; i < n_merges; i++) {
+                    const std::string word = gguf_get_arr_str(ctx, merges_keyidx, i);
+                    //GGML_ASSERT(unicode_cpts_from_utf8(word).size() > 0);
 
-            const int n_merges = gguf_get_arr_n(ctx, merges_keyidx);
-            for (int i = 0; i < n_merges; i++) {
-                const std::string word = gguf_get_arr_str(ctx, merges_keyidx, i);
-                //GGML_ASSERT(unicode_cpts_from_utf8(word).size() > 0);
+                    std::string first;
+                    std::string second;
 
-                std::string first;
-                std::string second;
+                    const size_t pos = word.find(' ', 1);
 
-                const size_t pos = word.find(' ', 1);
+                    if (pos != std::string::npos) {
+                        first  = word.substr(0, pos);
+                        second = word.substr(pos + 1);
+                    }
 
-                if (pos != std::string::npos) {
-                    first  = word.substr(0, pos);
-                    second = word.substr(pos + 1);
+                    bpe_ranks.emplace(std::make_pair(first, second), i);
                 }
-
-                bpe_ranks.emplace(std::make_pair(first, second), i);
             }
 
             // default special tokens
@@ -1794,7 +1833,7 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
                 const char * pc = (const char *) gguf_get_arr_data(ctx, precompiled_charsmap_keyidx);
                 precompiled_charsmap.assign(pc, pc + n_precompiled_charsmap);
 #if defined(__BYTE_ORDER__) && defined(__ORDER_BIG_ENDIAN__) && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
-                // correct endiannes of data in precompiled_charsmap binary blob
+                // correct endianness of data in precompiled_charsmap binary blob
                 uint32_t * xcda_blob_size = (uint32_t *) &precompiled_charsmap[0];
                 *xcda_blob_size = __builtin_bswap32(*xcda_blob_size);
                 assert(*xcda_blob_size + sizeof(uint32_t) < n_precompiled_charsmap);
@@ -1851,7 +1890,8 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
                     tokenizer_pre == "falcon-h1" ||
                     tokenizer_pre == "pixtral"  ||
                     tokenizer_pre == "midm-2.0" ||
-                    tokenizer_pre == "lfm2") {
+                    tokenizer_pre == "lfm2"     ||
+                    tokenizer_pre == "jina-v5-nano") {
                 pre_type = LLAMA_VOCAB_PRE_TYPE_LLAMA3;
                 ignore_merges = true;
                 add_bos = true;
@@ -1891,8 +1931,11 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
                     tokenizer_pre == "jina-v2-de" ||
                     tokenizer_pre == "a.x-4.0" ||
                     tokenizer_pre == "mellum"  ||
-                    tokenizer_pre == "modern-bert" ) {
+                    tokenizer_pre == "modern-bert") {
                 pre_type = LLAMA_VOCAB_PRE_TYPE_GPT2;
+            } else if (
+                    tokenizer_pre == "jais-2") {
+                pre_type = LLAMA_VOCAB_PRE_TYPE_JAIS2;
             } else if (
                     tokenizer_pre == "jina-v1-en" ||
                     tokenizer_pre == "jina-v2-code" ||
@@ -1912,6 +1955,10 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
                     tokenizer_pre == "kormo") {
                 pre_type = LLAMA_VOCAB_PRE_TYPE_QWEN2;
                 clean_spaces = false;
+            } else if (
+                    tokenizer_pre == "qwen35") {
+                pre_type = LLAMA_VOCAB_PRE_TYPE_QWEN35;
+                clean_spaces = false;
             } else if (
                 tokenizer_pre == "stablelm2") {
                 pre_type = LLAMA_VOCAB_PRE_TYPE_STABLELM2;
@@ -1965,6 +2012,9 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
             } else if (
                 tokenizer_pre == "exaone4") {
                 pre_type = LLAMA_VOCAB_PRE_TYPE_GPT2;
+            } else if (
+                tokenizer_pre == "exaone-moe") {
+                pre_type = LLAMA_VOCAB_PRE_TYPE_EXAONE_MOE;
             } else if (
                 tokenizer_pre == "chameleon") {
                 pre_type = LLAMA_VOCAB_PRE_TYPE_CHAMELEON;
@@ -1977,10 +2027,15 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
                 tokenizer_pre == "megrez") {
                 pre_type = LLAMA_VOCAB_PRE_TYPE_QWEN2;
             } else if (
-                    tokenizer_pre == "gpt-4o" ||
-                    tokenizer_pre == "llama4") {
+                tokenizer_pre == "gpt-4o" ||
+                tokenizer_pre == "llama4" ||
+                tokenizer_pre == "kanana2") {
                 pre_type = LLAMA_VOCAB_PRE_TYPE_GPT4O;
                 clean_spaces = false;
+            } else if (
+                tokenizer_pre == "tiny_aya") {
+                pre_type = LLAMA_VOCAB_PRE_TYPE_TINY_AYA;
+                clean_spaces = false;
             } else if (
                 tokenizer_pre == "superbpe") {
                 pre_type = LLAMA_VOCAB_PRE_TYPE_SUPERBPE;
@@ -2011,6 +2066,10 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
                 tokenizer_pre == "hunyuan-dense") {
                 pre_type = LLAMA_VOCAB_PRE_TYPE_HUNYUAN_DENSE;
                 clean_spaces = false;
+            } else if (
+                tokenizer_pre == "joyai-llm") {
+                pre_type = LLAMA_VOCAB_PRE_TYPE_JOYAI_LLM;
+                clean_spaces = false;
             } else if (
                 tokenizer_pre == "kimi-k2") {
                 pre_type = LLAMA_VOCAB_PRE_TYPE_KIMI_K2;
@@ -2216,6 +2275,7 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
                         || t.first == "<|end_of_text|>" // granite
                         || t.first == "<EOT>"
                         || t.first == "_<EOT>"
+                        || t.first == "[EOT]" // Kimi-K2
                         || t.first == "<｜end▁of▁sentence｜>" // DeepSeek
                         || t.first == "<end_of_utterance>" // smoldocling
                    ) {
@@ -2252,6 +2312,7 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
                         || t.first == "<PRE>"
                         || t.first == "▁<PRE>"          // CodeLlama
                         || t.first == "<|code_prefix|>" // GLM-4.5
+                        || t.first == "<|prefix|>"      // Falcon-H1-Tiny-Coder
                         ) {
                     special_fim_pre_id = t.second;
                     if ((attr & LLAMA_TOKEN_ATTR_CONTROL) == 0) {
@@ -2272,6 +2333,7 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
                         || t.first == "<SUF>"
                         || t.first == "▁<SUF>"         // CodeLlama
                         || t.first == "<|code_suffix|>" // GLM-4.5
+                        || t.first == "<|suffix|>"      // Falcon-H1-Tiny-Coder
                         ) {
                     special_fim_suf_id = t.second;
                     if ((attr & LLAMA_TOKEN_ATTR_CONTROL) == 0) {
@@ -2292,6 +2354,7 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
                         || t.first == "<MID>"
                         || t.first == "▁<MID>"         // CodeLlama
                         || t.first == "<|code_middle|>" // GLM-4.5
+                        || t.first == "<|middle|>"      // Falcon-H1-Tiny-Coder
                         ) {
                     special_fim_mid_id = t.second;
                     if ((attr & LLAMA_TOKEN_ATTR_CONTROL) == 0) {
@@ -2309,6 +2372,7 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
                         || t.first == "<fim-pad>"
                         || t.first == "<fim_pad>"   // Granite
                         || t.first == "<PAD>"
+                        || t.first == "[PAD]" // Kimi-K2
                         ) {
                     special_fim_pad_id = t.second;
                     if ((attr & LLAMA_TOKEN_ATTR_CONTROL) == 0) {
@@ -2380,7 +2444,7 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
 
         // maintain a list of tokens that cause end-of-generation
         // this is currently determined based on the token text, which is obviously not ideal
-        // ref: https://github.com/ggerganov/llama.cpp/issues/9606
+        // ref: https://github.com/ggml-org/llama.cpp/issues/9606
         special_eog_ids.clear();
 
         if (special_fim_pad_id != LLAMA_TOKEN_NULL && special_eog_ids.count(special_fim_pad_id) == 0) {
@@ -2408,9 +2472,12 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
                     || t.first == "<|calls|>"  // solar-open
                     || t.first == "<end_of_turn>"
                     || t.first == "<|endoftext|>"
+                    || t.first == "</s>"      // paddleocr
                     || t.first == "<|eom_id|>"
                     || t.first == "<EOT>"
                     || t.first == "_<EOT>"
+                    || t.first == "[EOT]" // Kimi-K2
+                    || t.first == "[EOS]" // Kimi-K2
                     || t.first == "<|end_of_text|>"
                     || t.first == "<end_of_utterance>" // smoldocling
                ) {
@@ -2436,7 +2503,10 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
             auto & attr = id_to_token[t.second].attr;
 
             if (t.first == "<|channel|>" || t.first == "<|message|>" || t.first == "<|start|>" || t.first == "<|constrain|>") {
-                attr = (llama_token_attr) (attr | LLAMA_TOKEN_ATTR_USER_DEFINED);
+                LLAMA_LOG_WARN("%s: setting token '%s' (%d) attribute to USER_DEFINED (%u), old attributes: %u\n",
+                        __func__, t.first.c_str(), t.second, LLAMA_TOKEN_ATTR_USER_DEFINED, attr);
+
+                attr = LLAMA_TOKEN_ATTR_USER_DEFINED;
             }
         }
 
@@ -2489,7 +2559,7 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
                 special_eog_ids.erase(end_id);
 
                 auto & attr = id_to_token[end_id].attr;
-                attr = (llama_token_attr) (attr | LLAMA_TOKEN_ATTR_USER_DEFINED);
+                attr = LLAMA_TOKEN_ATTR_USER_DEFINED;
 
                 LLAMA_LOG_WARN("%s: special_eog_ids contains both '<|return|>' and '<|call|>', or '<|calls|>' and '<|flush|>' tokens, removing '<|end|>' token from EOG list\n", __func__);
             }
@@ -3066,7 +3136,7 @@ std::vector<llama_token> llama_vocab::impl::tokenize(
 }
 
 int32_t llama_vocab::impl::token_to_piece(llama_token token, char * buf, int32_t length, int32_t lstrip, bool special) const {
-    // ref: https://github.com/ggerganov/llama.cpp/pull/7587#discussion_r1620983843
+    // ref: https://github.com/ggml-org/llama.cpp/pull/7587#discussion_r1620983843
     static const int attr_special = LLAMA_TOKEN_ATTR_UNKNOWN | LLAMA_TOKEN_ATTR_CONTROL;
     const llama_token_attr attr = token_get_attr(token);
     if (!special && (attr & attr_special)) {
@@ -3289,34 +3359,34 @@ int32_t llama_vocab::impl::detokenize(
 }
 
 void llama_vocab::impl::print_info() const {
-    LLAMA_LOG_INFO("%s: vocab type       = %s\n",     __func__, type_name().c_str());
-    LLAMA_LOG_INFO("%s: n_vocab          = %u\n",     __func__, vocab.n_tokens());
-    LLAMA_LOG_INFO("%s: n_merges         = %u\n",     __func__, (uint32_t) bpe_ranks.size());
+    LLAMA_LOG_INFO("%s: vocab type            = %s\n",     __func__, type_name().c_str());
+    LLAMA_LOG_INFO("%s: n_vocab               = %u\n",     __func__, vocab.n_tokens());
+    LLAMA_LOG_INFO("%s: n_merges              = %u\n",     __func__, (uint32_t) bpe_ranks.size());
 
     // special tokens
-    if (special_bos_id  != LLAMA_TOKEN_NULL)    { LLAMA_LOG_INFO( "%s: BOS token        = %d '%s'\n", __func__, special_bos_id,     id_to_token.at(special_bos_id).text.c_str() );  }
-    if (special_eos_id  != LLAMA_TOKEN_NULL)    { LLAMA_LOG_INFO( "%s: EOS token        = %d '%s'\n", __func__, special_eos_id,     id_to_token.at(special_eos_id).text.c_str() );  }
-    if (special_eot_id  != LLAMA_TOKEN_NULL)    { LLAMA_LOG_INFO( "%s: EOT token        = %d '%s'\n", __func__, special_eot_id,     id_to_token.at(special_eot_id).text.c_str() );  }
-    if (special_eom_id  != LLAMA_TOKEN_NULL)    { LLAMA_LOG_INFO( "%s: EOM token        = %d '%s'\n", __func__, special_eom_id,     id_to_token.at(special_eom_id).text.c_str() );  }
-    if (special_unk_id  != LLAMA_TOKEN_NULL)    { LLAMA_LOG_INFO( "%s: UNK token        = %d '%s'\n", __func__, special_unk_id,     id_to_token.at(special_unk_id).text.c_str() );  }
-    if (special_sep_id  != LLAMA_TOKEN_NULL)    { LLAMA_LOG_INFO( "%s: SEP token        = %d '%s'\n", __func__, special_sep_id,     id_to_token.at(special_sep_id).text.c_str() );  }
-    if (special_pad_id  != LLAMA_TOKEN_NULL)    { LLAMA_LOG_INFO( "%s: PAD token        = %d '%s'\n", __func__, special_pad_id,     id_to_token.at(special_pad_id).text.c_str() );  }
-    if (special_mask_id != LLAMA_TOKEN_NULL)    { LLAMA_LOG_INFO( "%s: MASK token       = %d '%s'\n", __func__, special_mask_id,    id_to_token.at(special_mask_id).text.c_str() ); }
-
-    if (linefeed_id != LLAMA_TOKEN_NULL)        { LLAMA_LOG_INFO( "%s: LF token         = %d '%s'\n", __func__, linefeed_id,        id_to_token.at(linefeed_id).text.c_str() ); }
-
-    if (special_fim_pre_id != LLAMA_TOKEN_NULL) { LLAMA_LOG_INFO( "%s: FIM PRE token    = %d '%s'\n", __func__, special_fim_pre_id, id_to_token.at(special_fim_pre_id).text.c_str() ); }
-    if (special_fim_suf_id != LLAMA_TOKEN_NULL) { LLAMA_LOG_INFO( "%s: FIM SUF token    = %d '%s'\n", __func__, special_fim_suf_id, id_to_token.at(special_fim_suf_id).text.c_str() ); }
-    if (special_fim_mid_id != LLAMA_TOKEN_NULL) { LLAMA_LOG_INFO( "%s: FIM MID token    = %d '%s'\n", __func__, special_fim_mid_id, id_to_token.at(special_fim_mid_id).text.c_str() ); }
-    if (special_fim_pad_id != LLAMA_TOKEN_NULL) { LLAMA_LOG_INFO( "%s: FIM PAD token    = %d '%s'\n", __func__, special_fim_pad_id, id_to_token.at(special_fim_pad_id).text.c_str() ); }
-    if (special_fim_rep_id != LLAMA_TOKEN_NULL) { LLAMA_LOG_INFO( "%s: FIM REP token    = %d '%s'\n", __func__, special_fim_rep_id, id_to_token.at(special_fim_rep_id).text.c_str() ); }
-    if (special_fim_sep_id != LLAMA_TOKEN_NULL) { LLAMA_LOG_INFO( "%s: FIM SEP token    = %d '%s'\n", __func__, special_fim_sep_id, id_to_token.at(special_fim_sep_id).text.c_str() ); }
+    if (special_bos_id  != LLAMA_TOKEN_NULL)    { LLAMA_LOG_INFO( "%s: BOS token             = %d '%s'\n", __func__, special_bos_id,     id_to_token.at(special_bos_id).text.c_str() );  }
+    if (special_eos_id  != LLAMA_TOKEN_NULL)    { LLAMA_LOG_INFO( "%s: EOS token             = %d '%s'\n", __func__, special_eos_id,     id_to_token.at(special_eos_id).text.c_str() );  }
+    if (special_eot_id  != LLAMA_TOKEN_NULL)    { LLAMA_LOG_INFO( "%s: EOT token             = %d '%s'\n", __func__, special_eot_id,     id_to_token.at(special_eot_id).text.c_str() );  }
+    if (special_eom_id  != LLAMA_TOKEN_NULL)    { LLAMA_LOG_INFO( "%s: EOM token             = %d '%s'\n", __func__, special_eom_id,     id_to_token.at(special_eom_id).text.c_str() );  }
+    if (special_unk_id  != LLAMA_TOKEN_NULL)    { LLAMA_LOG_INFO( "%s: UNK token             = %d '%s'\n", __func__, special_unk_id,     id_to_token.at(special_unk_id).text.c_str() );  }
+    if (special_sep_id  != LLAMA_TOKEN_NULL)    { LLAMA_LOG_INFO( "%s: SEP token             = %d '%s'\n", __func__, special_sep_id,     id_to_token.at(special_sep_id).text.c_str() );  }
+    if (special_pad_id  != LLAMA_TOKEN_NULL)    { LLAMA_LOG_INFO( "%s: PAD token             = %d '%s'\n", __func__, special_pad_id,     id_to_token.at(special_pad_id).text.c_str() );  }
+    if (special_mask_id != LLAMA_TOKEN_NULL)    { LLAMA_LOG_INFO( "%s: MASK token            = %d '%s'\n", __func__, special_mask_id,    id_to_token.at(special_mask_id).text.c_str() ); }
+
+    if (linefeed_id != LLAMA_TOKEN_NULL)        { LLAMA_LOG_INFO( "%s: LF token              = %d '%s'\n", __func__, linefeed_id,        id_to_token.at(linefeed_id).text.c_str() ); }
+
+    if (special_fim_pre_id != LLAMA_TOKEN_NULL) { LLAMA_LOG_INFO( "%s: FIM PRE token         = %d '%s'\n", __func__, special_fim_pre_id, id_to_token.at(special_fim_pre_id).text.c_str() ); }
+    if (special_fim_suf_id != LLAMA_TOKEN_NULL) { LLAMA_LOG_INFO( "%s: FIM SUF token         = %d '%s'\n", __func__, special_fim_suf_id, id_to_token.at(special_fim_suf_id).text.c_str() ); }
+    if (special_fim_mid_id != LLAMA_TOKEN_NULL) { LLAMA_LOG_INFO( "%s: FIM MID token         = %d '%s'\n", __func__, special_fim_mid_id, id_to_token.at(special_fim_mid_id).text.c_str() ); }
+    if (special_fim_pad_id != LLAMA_TOKEN_NULL) { LLAMA_LOG_INFO( "%s: FIM PAD token         = %d '%s'\n", __func__, special_fim_pad_id, id_to_token.at(special_fim_pad_id).text.c_str() ); }
+    if (special_fim_rep_id != LLAMA_TOKEN_NULL) { LLAMA_LOG_INFO( "%s: FIM REP token         = %d '%s'\n", __func__, special_fim_rep_id, id_to_token.at(special_fim_rep_id).text.c_str() ); }
+    if (special_fim_sep_id != LLAMA_TOKEN_NULL) { LLAMA_LOG_INFO( "%s: FIM SEP token         = %d '%s'\n", __func__, special_fim_sep_id, id_to_token.at(special_fim_sep_id).text.c_str() ); }
 
     for (const auto & id : special_eog_ids) {
-        LLAMA_LOG_INFO( "%s: EOG token        = %d '%s'\n", __func__, id, id_to_token.at(id).text.c_str() );
+        LLAMA_LOG_INFO( "%s: EOG token             = %d '%s'\n", __func__, id, id_to_token.at(id).text.c_str() );
     }
 
-    LLAMA_LOG_INFO("%s: max token length = %d\n", __func__, max_token_len);
+    LLAMA_LOG_INFO("%s: max token length      = %d\n", __func__, max_token_len);
 }
 
 llama_vocab::llama_vocab() : pimpl(new impl(*this)) {

data/ext/sources/examples/talk-llama/llama-vocab.h

@@ -53,6 +53,11 @@ enum llama_vocab_pre_type {
     LLAMA_VOCAB_PRE_TYPE_AFMOE           = 42,
     LLAMA_VOCAB_PRE_TYPE_SOLAR_OPEN      = 43,
     LLAMA_VOCAB_PRE_TYPE_YOUTU           = 44,
+    LLAMA_VOCAB_PRE_TYPE_EXAONE_MOE      = 45,
+    LLAMA_VOCAB_PRE_TYPE_QWEN35          = 46,
+    LLAMA_VOCAB_PRE_TYPE_TINY_AYA        = 47,
+    LLAMA_VOCAB_PRE_TYPE_JOYAI_LLM       = 48,
+    LLAMA_VOCAB_PRE_TYPE_JAIS2           = 49,
 };
 
 struct LLM_KV;

data/ext/sources/examples/talk-llama/llama.cpp

@@ -1,5 +1,6 @@
 #include "llama.h"
 
+#include "ggml-cpp.h"
 #include "llama-impl.h"
 
 #include "llama-chat.h"
@@ -12,6 +13,7 @@
 
 #include "ggml.h"
 #include "ggml-backend.h"
+#include "gguf.h"
 
 #include <algorithm>
 #include <cassert>
@@ -311,8 +313,12 @@ static void llama_params_fit_impl(
                             __func__, hp_nct, cparams->n_ctx, memory_reduction/MiB);
                     }
                 } else {
-                    LLAMA_LOG_INFO("%s: default model context size is %" PRIu32 " which is <= the min. context size of %" PRIu32 " -> no change\n",
-                        __func__, hp_nct, n_ctx_min);
+                    if (n_ctx_min == UINT32_MAX) {
+                        LLAMA_LOG_INFO("%s: user has requested full context size of %" PRIu32 " -> no change\n", __func__, hp_nct);
+                    } else {
+                        LLAMA_LOG_INFO("%s: default model context size is %" PRIu32 " which is <= the min. context size of %" PRIu32 " -> no change\n",
+                            __func__, hp_nct, n_ctx_min);
+                    }
                 }
             } else {
                 LLAMA_LOG_INFO("%s: context size set by user to %" PRIu32 " -> no change\n", __func__, cparams->n_ctx);
@@ -821,7 +827,8 @@ int64_t llama_time_us(void) {
 }
 
 // Returns 0 on success, -1 on error, and -2 on cancellation via llama_progress_callback
-static int llama_model_load(const std::string & fname, std::vector<std::string> & splits, llama_model & model, llama_model_params & params) {
+static int llama_model_load(struct gguf_context * metadata, llama_model_set_tensor_data_t set_tensor_data, void * set_tensor_data_ud,
+        const std::string & fname, std::vector<std::string> & splits, llama_model & model, llama_model_params & params) {
     // loading time will be recalculated after the first eval, so
     // we take page faults deferred by mmap() into consideration
     model.t_load_us = 0;
@@ -830,7 +837,8 @@ static int llama_model_load(const std::string & fname, std::vector<std::string>
     model.t_start_us = tm.t_start_us;
 
     try {
-        llama_model_loader ml(fname, splits, params.use_mmap, params.use_direct_io, params.check_tensors, params.no_alloc, params.kv_overrides, params.tensor_buft_overrides);
+        llama_model_loader ml(metadata, set_tensor_data, set_tensor_data_ud, fname, splits, params.use_mmap, params.use_direct_io,
+            params.check_tensors, params.no_alloc, params.kv_overrides, params.tensor_buft_overrides);
 
         ml.print_info();
 
@@ -876,9 +884,13 @@ static int llama_model_load(const std::string & fname, std::vector<std::string>
 }
 
 static struct llama_model * llama_model_load_from_file_impl(
+        struct gguf_context * metadata,
+        llama_model_set_tensor_data_t set_tensor_data,
+        void * set_tensor_data_ud,
         const std::string & path_model,
         std::vector<std::string> & splits,
         struct llama_model_params params) {
+    GGML_ASSERT((metadata == nullptr) != path_model.empty() && "exactly one out of metadata and path_model needs to be defined");
     ggml_time_init();
 
     if (!params.vocab_only && ggml_backend_reg_count() == 0) {
@@ -999,7 +1011,7 @@ static struct llama_model * llama_model_load_from_file_impl(
                 props.memory_free/1024/1024);
     }
 
-    const int status = llama_model_load(path_model, splits, *model, params);
+    const int status = llama_model_load(metadata, set_tensor_data, set_tensor_data_ud, path_model, splits, *model, params);
     GGML_ASSERT(status <= 0);
     if (status < 0) {
         if (status == -1) {
@@ -1015,6 +1027,18 @@ static struct llama_model * llama_model_load_from_file_impl(
     return model;
 }
 
+struct llama_model * llama_model_init_from_user(
+        struct gguf_context * metadata,
+        llama_model_set_tensor_data_t set_tensor_data,
+        void * set_tensor_data_ud,
+        struct llama_model_params params) {
+    GGML_ASSERT(metadata != nullptr);
+    std::string path_model;
+    std::vector<std::string> splits = {};
+    params.use_mmap = false;
+    params.use_extra_bufts = false;
+    return llama_model_load_from_file_impl(metadata, set_tensor_data, set_tensor_data_ud, path_model, splits, params);
+}
 // deprecated
 struct llama_model * llama_load_model_from_file(
         const char * path_model,
@@ -1026,7 +1050,7 @@ struct llama_model * llama_model_load_from_file(
         const char * path_model,
         struct llama_model_params params) {
     std::vector<std::string> splits = {};
-    return llama_model_load_from_file_impl(path_model, splits, params);
+    return llama_model_load_from_file_impl(nullptr, nullptr, nullptr, path_model, splits, params);
 }
 
 struct llama_model * llama_model_load_from_splits(
@@ -1042,11 +1066,11 @@ struct llama_model * llama_model_load_from_splits(
     for (size_t i = 0; i < n_paths; ++i) {
         splits.push_back(paths[i]);
     }
-    return llama_model_load_from_file_impl(splits.front(), splits, params);
+    return llama_model_load_from_file_impl(nullptr, nullptr, nullptr, splits.front(), splits, params);
 }
 
 void llama_model_save_to_file(const struct llama_model * model, const char * path_model) {
-    llama_model_saver ms(*model);
+    llama_model_saver ms(model);
     ms.add_kv_from_model();
     ms.add_tensors_from_model();
     ms.save(path_model);
@@ -1091,25 +1115,55 @@ int32_t llama_chat_apply_template(
 // model split
 //
 
-int llama_split_path(char * split_path, size_t maxlen, const char * path_prefix, int split_no, int split_count) {
+int32_t llama_split_path(
+    char * split_path,
+    size_t maxlen,
+    const char * path_prefix,
+    int32_t split_no,
+    int32_t split_count) {
+
     static const char * const SPLIT_PATH_FORMAT = "%s-%05d-of-%05d.gguf";
-    if (snprintf(split_path, maxlen, SPLIT_PATH_FORMAT, path_prefix, split_no + 1, split_count)) {
-        return strlen(split_path);
+
+    const int written = snprintf(
+        split_path,
+        maxlen,
+        SPLIT_PATH_FORMAT,
+        path_prefix,
+        split_no + 1,
+        split_count
+    );
+
+    if (written < 0 || (size_t) written >= maxlen) {
+        return 0;
     }
-    return 0;
+
+    return (int32_t) written;
 }
 
-int llama_split_prefix(char * split_prefix, size_t maxlen, const char * split_path, int split_no, int split_count) {
-    std::string str_split_path(split_path);
+int32_t llama_split_prefix(
+    char * split_prefix,
+    size_t maxlen,
+    const char * split_path,
+    int32_t split_no,
+    int32_t split_count) {
+
+    const std::string str_split_path(split_path);
+
     char postfix[32];
-    snprintf(postfix, 32, "-%05d-of-%05d.gguf", split_no + 1, split_count);
-    std::string str_postfix(postfix);
-
-    // check if split_prefix ends with postfix
-    int size_prefix = str_split_path.size() - str_postfix.size();
-    if (size_prefix > 0 && str_split_path.find(str_postfix, size_prefix) != std::string::npos) {
-        snprintf(split_prefix, std::min((size_t) size_prefix + 1, maxlen), "%s", split_path);
-        return size_prefix;
+    snprintf(postfix, sizeof(postfix), "-%05d-of-%05d.gguf", split_no + 1, split_count);
+
+    const std::string str_postfix(postfix);
+    if (str_split_path.size() <= str_postfix.size()) {
+        return 0;
+    }
+
+    const size_t size_prefix = str_split_path.size() - str_postfix.size();
+
+    if (str_split_path.compare(size_prefix, std::string::npos, str_postfix) == 0) {
+        const size_t copy_len = std::min(size_prefix + 1, maxlen);
+        snprintf(split_prefix, copy_len, "%s", split_path);
+
+        return (int32_t) size_prefix;
     }
 
     return 0;