whispercpp 1.3.2 → 1.3.3

data/ext/sources/examples/server/server.cpp

@@ -5,6 +5,7 @@
 #include "httplib.h"
 #include "json.hpp"
 
+#include <cfloat>
 #include <chrono>
 #include <cmath>
 #include <cstdio>
@@ -13,10 +14,23 @@
 #include <string>
 #include <thread>
 #include <vector>
+#include <memory>
+#include <csignal>
+#include <atomic>
+#include <functional>
+#include <cstdlib>
+#if defined (_WIN32)
+#include <windows.h>
+#endif
 
 using namespace httplib;
 using json = nlohmann::ordered_json;
 
+enum server_state {
+    SERVER_STATE_LOADING_MODEL,  // Server is starting up, model not fully loaded yet
+    SERVER_STATE_READY,          // Server is ready and model is loaded
+};
+
 namespace {
 
 // output formats
@@ -26,6 +40,20 @@ const std::string srt_format    = "srt";
 const std::string vjson_format  = "verbose_json";
 const std::string vtt_format    = "vtt";
 
+std::function<void(int)> shutdown_handler;
+std::atomic_flag is_terminating = ATOMIC_FLAG_INIT;
+
+inline void signal_handler(int signal) {
+    if (is_terminating.test_and_set()) {
+        // in case it hangs, we can force terminate the server by hitting Ctrl+C twice
+        // this is for better developer experience, we can remove when the server is stable enough
+        fprintf(stderr, "Received second interrupt, terminating immediately.\n");
+        exit(1);
+    }
+
+    shutdown_handler(signal);
+}
+
 struct server_params
 {
     std::string hostname = "127.0.0.1";
@@ -90,6 +118,16 @@ struct whisper_params {
     std::string openvino_encode_device = "CPU";
 
     std::string dtw = "";
+
+    // Voice Activity Detection (VAD) parameters
+    bool        vad           = false;
+    std::string vad_model     = "";
+    float       vad_threshold = 0.5f;
+    int         vad_min_speech_duration_ms = 250;
+    int         vad_min_silence_duration_ms = 100;
+    float       vad_max_speech_duration_s = FLT_MAX;
+    int         vad_speech_pad_ms = 30;
+    float       vad_samples_overlap = 0.1f;
 };
 
 void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params & params, const server_params& sparams) {
@@ -140,6 +178,18 @@ void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params & para
     fprintf(stderr, "  -nc,       --no-context        [%-7s] do not use previous audio context\n", params.no_context ? "true" : "false");
     fprintf(stderr, "  -ng,       --no-gpu            [%-7s] do not use gpu\n", params.use_gpu ? "false" : "true");
     fprintf(stderr, "  -fa,       --flash-attn        [%-7s] flash attention\n", params.flash_attn ? "true" : "false");
+    // Voice Activity Detection (VAD) parameters
+    fprintf(stderr, "\nVoice Activity Detection (VAD) options:\n");
+    fprintf(stderr, "             --vad                           [%-7s] enable Voice Activity Detection (VAD)\n",            params.vad ? "true" : "false");
+    fprintf(stderr, "  -vm FNAME, --vad-model FNAME               [%-7s] VAD model path\n",                                   params.vad_model.c_str());
+    fprintf(stderr, "  -vt N,     --vad-threshold N               [%-7.2f] VAD threshold for speech recognition\n",           params.vad_threshold);
+    fprintf(stderr, "  -vspd N,   --vad-min-speech-duration-ms  N [%-7d] VAD min speech duration (0.0-1.0)\n",                params.vad_min_speech_duration_ms);
+    fprintf(stderr, "  -vsd N,    --vad-min-silence-duration-ms N [%-7d] VAD min silence duration (to split segments)\n",      params.vad_min_silence_duration_ms);
+    fprintf(stderr, "  -vmsd N,   --vad-max-speech-duration-s   N [%-7s] VAD max speech duration (auto-split longer)\n",      params.vad_max_speech_duration_s == FLT_MAX ?
+                                                                                                                                  std::string("FLT_MAX").c_str() :
+                                                                                                                                  std::to_string(params.vad_max_speech_duration_s).c_str());
+    fprintf(stderr, "  -vp N,     --vad-speech-pad-ms           N [%-7d] VAD speech padding (extend segments)\n",             params.vad_speech_pad_ms);
+    fprintf(stderr, "  -vo N,     --vad-samples-overlap         N [%-7.2f] VAD samples overlap (seconds between segments)\n", params.vad_samples_overlap);
     fprintf(stderr, "\n");
 }
 
@@ -195,6 +245,16 @@ bool whisper_params_parse(int argc, char ** argv, whisper_params & params, serve
         else if (                  arg == "--request-path")    { sparams.request_path = argv[++i]; }
         else if (                  arg == "--inference-path")  { sparams.inference_path = argv[++i]; }
         else if (                  arg == "--convert")         { sparams.ffmpeg_converter     = true; }
+
+        // Voice Activity Detection (VAD)
+        else if (                  arg == "--vad")                         { params.vad                         = true; }
+        else if (arg == "-vm"   || arg == "--vad-model")                   { params.vad_model                   = argv[++i]; }
+        else if (arg == "-vt"   || arg == "--vad-threshold")               { params.vad_threshold               = std::stof(argv[++i]); }
+        else if (arg == "-vspd" || arg == "--vad-min-speech-duration-ms")  { params.vad_min_speech_duration_ms  = std::stoi(argv[++i]); }
+        else if (arg == "-vsd"  || arg == "--vad-min-silence-duration-ms") { params.vad_min_speech_duration_ms  = std::stoi(argv[++i]); }
+        else if (arg == "-vmsd" || arg == "--vad-max-speech-duration-s")   { params.vad_max_speech_duration_s   = std::stof(argv[++i]); }
+        else if (arg == "-vp"   || arg == "--vad-speech-pad-ms")           { params.vad_speech_pad_ms           = std::stoi(argv[++i]); }
+        else if (arg == "-vo"   || arg == "--vad-samples-overlap")         { params.vad_samples_overlap         = std::stof(argv[++i]); }
         else {
             fprintf(stderr, "error: unknown argument: %s\n", arg.c_str());
             whisper_print_usage(argc, argv, params, sparams);
@@ -511,11 +571,41 @@ void get_req_parameters(const Request & req, whisper_params & params)
     {
         params.no_context = parse_str_to_bool(req.get_file_value("no_context").content);
     }
+    if (req.has_file("vad"))
+    {
+        params.vad = parse_str_to_bool(req.get_file_value("vad").content);
+    }
+    if (req.has_file("vad_threshold"))
+    {
+        params.vad_threshold = std::stof(req.get_file_value("vad_threshold").content);
+    }
+    if (req.has_file("vad_min_speech_duration_ms"))
+    {
+        params.vad_min_speech_duration_ms = std::stof(req.get_file_value("vad_min_speech_duration_ms").content);
+    }
+    if (req.has_file("vad_min_silence_duration_ms"))
+    {
+        params.vad_min_silence_duration_ms = std::stof(req.get_file_value("vad_min_silence_duration_ms").content);
+    }
+    if (req.has_file("vad_max_speech_duration_s"))
+    {
+        params.vad_max_speech_duration_s = std::stof(req.get_file_value("vad_max_speech_duration_s").content);
+    }
+    if (req.has_file("vad_speech_pad_ms"))
+    {
+        params.vad_speech_pad_ms = std::stoi(req.get_file_value("vad_speech_pad_ms").content);
+    }
+    if (req.has_file("vad_samples_overlap"))
+    {
+        params.vad_samples_overlap = std::stof(req.get_file_value("vad_samples_overlap").content);
+    }
 }
 
 }  // namespace
 
 int main(int argc, char ** argv) {
+    ggml_backend_load_all();
+
     whisper_params params;
     server_params sparams;
 
@@ -584,13 +674,19 @@ int main(int argc, char ** argv) {
         if (params.dtw == "large.v3") {
             cparams.dtw_aheads_preset = WHISPER_AHEADS_LARGE_V3;
         }
-
+        if (params.dtw == "large.v3.turbo") { 
+            cparams.dtw_aheads_preset = WHISPER_AHEADS_LARGE_V3_TURBO;
+        }
+        
         if (cparams.dtw_aheads_preset == WHISPER_AHEADS_NONE) {
             fprintf(stderr, "error: unknown DTW preset '%s'\n", params.dtw.c_str());
             return 3;
         }
     }
 
+    std::unique_ptr<httplib::Server> svr = std::make_unique<httplib::Server>();
+    std::atomic<server_state> state{SERVER_STATE_LOADING_MODEL};
+
     struct whisper_context * ctx = whisper_init_from_file_with_params(params.model.c_str(), cparams);
 
     if (ctx == nullptr) {
@@ -600,9 +696,10 @@ int main(int argc, char ** argv) {
 
     // initialize openvino encoder. this has no effect on whisper.cpp builds that don't have OpenVINO configured
     whisper_ctx_init_openvino_encoder(ctx, nullptr, params.openvino_encode_device.c_str(), nullptr);
+    state.store(SERVER_STATE_READY);
+
 
-    Server svr;
-    svr.set_default_headers({{"Server", "whisper.cpp"},
+    svr->set_default_headers({{"Server", "whisper.cpp"},
                              {"Access-Control-Allow-Origin", "*"},
                              {"Access-Control-Allow-Headers", "content-type, authorization"}});
 
@@ -681,15 +778,15 @@ int main(int argc, char ** argv) {
     whisper_params default_params = params;
 
     // this is only called if no index.html is found in the public --path
-    svr.Get(sparams.request_path + "/", [&default_content](const Request &, Response &res){
+    svr->Get(sparams.request_path + "/", [&](const Request &, Response &res){
         res.set_content(default_content, "text/html");
         return false;
     });
 
-    svr.Options(sparams.request_path + sparams.inference_path, [&](const Request &, Response &){
+    svr->Options(sparams.request_path + sparams.inference_path, [&](const Request &, Response &){
     });
 
-    svr.Post(sparams.request_path + sparams.inference_path, [&](const Request &req, Response &res){
+    svr->Post(sparams.request_path + sparams.inference_path, [&](const Request &req, Response &res){
         // acquire whisper model mutex lock
         std::lock_guard<std::mutex> lock(whisper_mutex);
 
@@ -827,6 +924,16 @@ int main(int argc, char ** argv) {
 
             wparams.suppress_nst     = params.suppress_nst;
 
+            wparams.vad              = params.vad;
+            wparams.vad_model_path   = params.vad_model.c_str();
+
+            wparams.vad_params.threshold               = params.vad_threshold;
+            wparams.vad_params.min_speech_duration_ms  = params.vad_min_speech_duration_ms;
+            wparams.vad_params.min_silence_duration_ms = params.vad_min_silence_duration_ms;
+            wparams.vad_params.max_speech_duration_s   = params.vad_max_speech_duration_s;
+            wparams.vad_params.speech_pad_ms           = params.vad_speech_pad_ms;
+            wparams.vad_params.samples_overlap         = params.vad_samples_overlap;
+
             whisper_print_user_data user_data = { &params, &pcmf32s, 0 };
 
             // this callback is called on each new segment
@@ -995,8 +1102,9 @@ int main(int argc, char ** argv) {
         // reset params to their defaults
         params = default_params;
     });
-    svr.Post(sparams.request_path + "/load", [&](const Request &req, Response &res){
+    svr->Post(sparams.request_path + "/load", [&](const Request &req, Response &res){
         std::lock_guard<std::mutex> lock(whisper_mutex);
+        state.store(SERVER_STATE_LOADING_MODEL);
         if (!req.has_file("model"))
         {
             fprintf(stderr, "error: no 'model' field in the request\n");
@@ -1028,18 +1136,25 @@ int main(int argc, char ** argv) {
         // initialize openvino encoder. this has no effect on whisper.cpp builds that don't have OpenVINO configured
         whisper_ctx_init_openvino_encoder(ctx, nullptr, params.openvino_encode_device.c_str(), nullptr);
 
+        state.store(SERVER_STATE_READY);
         const std::string success = "Load was successful!";
         res.set_content(success, "application/text");
 
         // check if the model is in the file system
     });
 
-    svr.Get(sparams.request_path + "/health", [&](const Request &, Response &res){
-        const std::string health_response = "{\"status\":\"ok\"}";
-        res.set_content(health_response, "application/json");
+    svr->Get(sparams.request_path + "/health", [&](const Request &, Response &res){
+        server_state current_state = state.load();
+        if (current_state == SERVER_STATE_READY) {
+            const std::string health_response = "{\"status\":\"ok\"}";
+            res.set_content(health_response, "application/json");
+        } else {
+            res.set_content("{\"status\":\"loading model\"}", "application/json");
+            res.status = 503;
+        }
     });
 
-    svr.set_exception_handler([](const Request &, Response &res, std::exception_ptr ep) {
+    svr->set_exception_handler([](const Request &, Response &res, std::exception_ptr ep) {
         const char fmt[] = "500 Internal Server Error\n%s";
         char buf[BUFSIZ];
         try {
@@ -1053,7 +1168,7 @@ int main(int argc, char ** argv) {
         res.status = 500;
     });
 
-    svr.set_error_handler([](const Request &req, Response &res) {
+    svr->set_error_handler([](const Request &req, Response &res) {
         if (res.status == 400) {
             res.set_content("Invalid request", "text/plain");
         } else if (res.status != 500) {
@@ -1063,10 +1178,10 @@ int main(int argc, char ** argv) {
     });
 
     // set timeouts and change hostname and port
-    svr.set_read_timeout(sparams.read_timeout);
-    svr.set_write_timeout(sparams.write_timeout);
+    svr->set_read_timeout(sparams.read_timeout);
+    svr->set_write_timeout(sparams.write_timeout);
 
-    if (!svr.bind_to_port(sparams.hostname, sparams.port))
+    if (!svr->bind_to_port(sparams.hostname, sparams.port))
     {
         fprintf(stderr, "\ncouldn't bind to server socket: hostname=%s port=%d\n\n",
                 sparams.hostname.c_str(), sparams.port);
@@ -1074,18 +1189,50 @@ int main(int argc, char ** argv) {
     }
 
     // Set the base directory for serving static files
-    svr.set_base_dir(sparams.public_path);
+    svr->set_base_dir(sparams.public_path);
 
     // to make it ctrl+clickable:
     printf("\nwhisper server listening at http://%s:%d\n\n", sparams.hostname.c_str(), sparams.port);
 
-    if (!svr.listen_after_bind())
-    {
-        return 1;
-    }
+    shutdown_handler = [&](int signal) {
+        printf("\nCaught signal %d, shutting down gracefully...\n", signal);
+        if (svr) {
+            svr->stop();
+        }
+    };
+
+#if defined (__unix__) || (defined (__APPLE__) && defined (__MACH__))
+    struct sigaction sigint_action;
+    sigint_action.sa_handler = signal_handler;
+    sigemptyset (&sigint_action.sa_mask);
+    sigint_action.sa_flags = 0;
+    sigaction(SIGINT, &sigint_action, NULL);
+    sigaction(SIGTERM, &sigint_action, NULL);
+#elif defined (_WIN32)
+    auto console_ctrl_handler = +[](DWORD ctrl_type) -> BOOL {
+        return (ctrl_type == CTRL_C_EVENT) ? (signal_handler(SIGINT), true) : false;
+    };
+    SetConsoleCtrlHandler(reinterpret_cast<PHANDLER_ROUTINE>(console_ctrl_handler), true);
+#endif
+
+    // clean up function, to be called before exit
+    auto clean_up = [&]() {
+        whisper_print_timings(ctx);
+        whisper_free(ctx);
+    };
+
+    std::thread t([&] {
+        if (!svr->listen_after_bind()) {
+            fprintf(stderr, "error: server listen failed\n");
+        }
+    });
+
+    svr->wait_until_ready();
+
+    t.join();
+
 
-    whisper_print_timings(ctx);
-    whisper_free(ctx);
+    clean_up();
 
     return 0;
 }

data/ext/sources/examples/stream/stream.cpp

@@ -116,6 +116,8 @@ void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params & para
 }
 
 int main(int argc, char ** argv) {
+    ggml_backend_load_all();
+
     whisper_params params;
 
     if (whisper_params_parse(argc, argv, params) == false) {
@@ -161,6 +163,10 @@ int main(int argc, char ** argv) {
     cparams.flash_attn = params.flash_attn;
 
     struct whisper_context * ctx = whisper_init_from_file_with_params(params.model.c_str(), cparams);
+    if (ctx == nullptr) {
+        fprintf(stderr, "error: failed to initialize whisper context\n");
+        return 2;
+    }
 
     std::vector<float> pcmf32    (n_samples_30s, 0.0f);
     std::vector<float> pcmf32_old;

data/ext/sources/examples/talk-llama/CMakeLists.txt

@@ -16,7 +16,10 @@ if (WHISPER_SDL2)
         llama-hparams.cpp
         llama-impl.cpp
         llama-io.cpp
-        llama-kv-cache.cpp
+        llama-kv-cache-unified.cpp
+        llama-kv-cache-unified-iswa.cpp
+        llama-memory-recurrent.cpp
+        llama-memory-hybrid.cpp
         llama-memory.cpp
         llama-mmap.cpp
         llama-model-loader.cpp

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

@@ -20,6 +20,7 @@ static const std::map<llm_arch, const char *> LLM_ARCH_NAMES = {
     { LLM_ARCH_BERT,             "bert"             },
     { LLM_ARCH_NOMIC_BERT,       "nomic-bert"       },
     { LLM_ARCH_NOMIC_BERT_MOE,   "nomic-bert-moe"   },
+    { LLM_ARCH_NEO_BERT,         "neo-bert"         },
     { LLM_ARCH_JINA_BERT_V2,     "jina-bert-v2"     },
     { LLM_ARCH_BLOOM,            "bloom"            },
     { LLM_ARCH_STABLELM,         "stablelm"         },
@@ -41,6 +42,7 @@ static const std::map<llm_arch, const char *> LLM_ARCH_NAMES = {
     { LLM_ARCH_GEMMA,            "gemma"            },
     { LLM_ARCH_GEMMA2,           "gemma2"           },
     { LLM_ARCH_GEMMA3,           "gemma3"           },
+    { LLM_ARCH_GEMMA3N,          "gemma3n"          },
     { LLM_ARCH_STARCODER2,       "starcoder2"       },
     { LLM_ARCH_MAMBA,            "mamba"            },
     { LLM_ARCH_XVERSE,           "xverse"           },
@@ -72,6 +74,9 @@ static const std::map<llm_arch, const char *> LLM_ARCH_NAMES = {
     { LLM_ARCH_WAVTOKENIZER_DEC, "wavtokenizer-dec" },
     { LLM_ARCH_PLM,              "plm"              },
     { LLM_ARCH_BAILINGMOE,       "bailingmoe"       },
+    { LLM_ARCH_DOTS1,            "dots1"            },
+    { LLM_ARCH_ARCEE,            "arcee"            },
+    { LLM_ARCH_ERNIE4_5,         "ernie4_5"         },
     { LLM_ARCH_UNKNOWN,          "(unknown)"        },
 };
 
@@ -144,6 +149,7 @@ static const std::map<llm_kv, const char *> LLM_KV_NAMES = {
     { LLM_KV_ATTENTION_SCALE,                        "%s.attention.scale"                        },
     { LLM_KV_ATTENTION_KEY_LENGTH_MLA,               "%s.attention.key_length_mla"               },
     { LLM_KV_ATTENTION_VALUE_LENGTH_MLA,             "%s.attention.value_length_mla"             },
+    { LLM_KV_ATTENTION_LAYER_INDICES,                "%s.attention.layer_indices"                },
 
     { LLM_KV_ROPE_DIMENSION_COUNT,      "%s.rope.dimension_count"                 },
     { LLM_KV_ROPE_DIMENSION_SECTIONS,   "%s.rope.dimension_sections"              },
@@ -174,6 +180,8 @@ static const std::map<llm_kv, const char *> LLM_KV_NAMES = {
     { LLM_KV_CONVNEXT_EMBEDDING_LENGTH, "%s.convnext.embedding_length" },
     { LLM_KV_CONVNEXT_BLOCK_COUNT,      "%s.convnext.block_count"      },
 
+    { LLM_KV_CLASSIFIER_OUTPUT_LABELS, "%s.classifier.output_labels" },
+
     { LLM_KV_TOKENIZER_MODEL,                "tokenizer.ggml.model"                    },
     { LLM_KV_TOKENIZER_PRE,                  "tokenizer.ggml.pre"                      },
     { LLM_KV_TOKENIZER_LIST,                 "tokenizer.ggml.tokens"                   },
@@ -192,13 +200,13 @@ static const std::map<llm_kv, const char *> LLM_KV_NAMES = {
     { LLM_KV_TOKENIZER_MASK_ID,              "tokenizer.ggml.mask_token_id"            },
     { LLM_KV_TOKENIZER_ADD_BOS,              "tokenizer.ggml.add_bos_token"            },
     { LLM_KV_TOKENIZER_ADD_EOS,              "tokenizer.ggml.add_eos_token"            },
+    { LLM_KV_TOKENIZER_ADD_SEP,              "tokenizer.ggml.add_sep_token"            },
     { LLM_KV_TOKENIZER_ADD_PREFIX,           "tokenizer.ggml.add_space_prefix"         },
     { LLM_KV_TOKENIZER_REMOVE_EXTRA_WS,      "tokenizer.ggml.remove_extra_whitespaces" },
     { LLM_KV_TOKENIZER_PRECOMPILED_CHARSMAP, "tokenizer.ggml.precompiled_charsmap"     },
     { LLM_KV_TOKENIZER_HF_JSON,              "tokenizer.huggingface.json"              },
     { LLM_KV_TOKENIZER_RWKV,                 "tokenizer.rwkv.world"                    },
     { LLM_KV_TOKENIZER_CHAT_TEMPLATE,        "tokenizer.chat_template"                 },
-    { LLM_KV_TOKENIZER_CHAT_TEMPLATE_N,      "tokenizer.chat_template.%s"              },
     { LLM_KV_TOKENIZER_FIM_PRE_ID,           "tokenizer.ggml.fim_pre_token_id"         },
     { LLM_KV_TOKENIZER_FIM_SUF_ID,           "tokenizer.ggml.fim_suf_token_id"         },
     { LLM_KV_TOKENIZER_FIM_MID_ID,           "tokenizer.ggml.fim_mid_token_id"         },
@@ -242,6 +250,24 @@ static const std::map<llm_arch, std::map<llm_tensor, const char *>> LLM_TENSOR_N
             { LLM_TENSOR_FFN_UP_EXPS,     "blk.%d.ffn_up_exps" },
         },
     },
+    {
+        LLM_ARCH_ARCEE,
+        {
+            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
+            { LLM_TENSOR_OUTPUT_NORM,     "output_norm" },
+            { LLM_TENSOR_OUTPUT,          "output" },
+            { LLM_TENSOR_ROPE_FREQS,      "rope_freqs" },
+            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
+            { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
+            { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
+            { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
+            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
+            { LLM_TENSOR_ATTN_ROT_EMBD,   "blk.%d.attn_rot_embd" },
+            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
+            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
+            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
+        },
+    },
     {
         LLM_ARCH_LLAMA4,
         {
@@ -448,6 +474,7 @@ static const std::map<llm_arch, std::map<llm_tensor, const char *>> LLM_TENSOR_N
             { LLM_TENSOR_TOKEN_TYPES,     "token_types" },
             { LLM_TENSOR_POS_EMBD,        "position_embd" },
             { LLM_TENSOR_ATTN_OUT_NORM,   "blk.%d.attn_output_norm" },
+            { LLM_TENSOR_ATTN_QKV,        "blk.%d.attn_qkv" },
             { LLM_TENSOR_ATTN_Q,          "blk.%d.attn_q" },
             { LLM_TENSOR_ATTN_K,          "blk.%d.attn_k" },
             { LLM_TENSOR_ATTN_V,          "blk.%d.attn_v" },
@@ -492,6 +519,21 @@ static const std::map<llm_arch, std::map<llm_tensor, const char *>> LLM_TENSOR_N
             { LLM_TENSOR_FFN_UP_EXPS,     "blk.%d.ffn_up_exps" },
         },
     },
+    {
+        LLM_ARCH_NEO_BERT,
+        {
+            { LLM_TENSOR_TOKEN_EMBD,      "token_embd" },
+            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm" },
+            { LLM_TENSOR_ATTN_QKV,        "blk.%d.attn_qkv" },
+            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output" },
+            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm" },
+            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down" },
+            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
+            { LLM_TENSOR_ENC_OUTPUT_NORM, "enc.output_norm" },
+            { LLM_TENSOR_CLS,             "cls" },
+            { LLM_TENSOR_CLS_OUT,         "cls.output" },
+        },
+    },
     {
         LLM_ARCH_JINA_BERT_V2,
         {
@@ -892,6 +934,42 @@ static const std::map<llm_arch, std::map<llm_tensor, const char *>> LLM_TENSOR_N
             { LLM_TENSOR_FFN_POST_NORM,   "blk.%d.post_ffw_norm" },
         },
     },
+    {
+        LLM_ARCH_GEMMA3N,
+        {
+            { LLM_TENSOR_TOKEN_EMBD,           "token_embd" },
+            { LLM_TENSOR_OUTPUT_NORM,          "output_norm" },
+            { LLM_TENSOR_ATTN_NORM,            "blk.%d.attn_norm" },
+            { LLM_TENSOR_ATTN_Q,               "blk.%d.attn_q" },
+            { LLM_TENSOR_ATTN_Q_NORM,          "blk.%d.attn_q_norm" },
+            { LLM_TENSOR_ATTN_K,               "blk.%d.attn_k" },
+            { LLM_TENSOR_ATTN_K_NORM,          "blk.%d.attn_k_norm" },
+            { LLM_TENSOR_ATTN_V,               "blk.%d.attn_v" },
+            { LLM_TENSOR_ATTN_OUT,             "blk.%d.attn_output" },
+            { LLM_TENSOR_ATTN_POST_NORM,       "blk.%d.post_attention_norm" },
+            { LLM_TENSOR_FFN_NORM,             "blk.%d.ffn_norm" },
+            { LLM_TENSOR_FFN_GATE,             "blk.%d.ffn_gate" },
+            { LLM_TENSOR_FFN_DOWN,             "blk.%d.ffn_down" },
+            { LLM_TENSOR_FFN_UP,               "blk.%d.ffn_up" },
+            { LLM_TENSOR_FFN_POST_NORM,        "blk.%d.post_ffw_norm" },
+            { LLM_TENSOR_PER_LAYER_TOKEN_EMBD, "per_layer_token_embd" },
+            { LLM_TENSOR_PER_LAYER_MODEL_PROJ, "per_layer_model_proj" },
+            { LLM_TENSOR_PER_LAYER_PROJ_NORM,  "per_layer_proj_norm" },
+            { LLM_TENSOR_ALTUP_UNEMBD_PROJ,    "altup_unembd_proj" },
+            { LLM_TENSOR_ALTUP_PROJ,           "altup_proj" },
+            { LLM_TENSOR_PER_LAYER_INP_GATE,   "blk.%d.inp_gate" },
+            { LLM_TENSOR_PER_LAYER_PROJ,       "blk.%d.proj" },
+            { LLM_TENSOR_PER_LAYER_POST_NORM,  "blk.%d.post_norm" },
+            { LLM_TENSOR_ALTUP_CORRECT_COEF,   "blk.%d.altup_correct_coef" },
+            { LLM_TENSOR_ALTUP_CORRECT_SCALE,  "blk.%d.altup_correct_scale" },
+            { LLM_TENSOR_ALTUP_PREDICT_COEF,   "blk.%d.altup_predict_coef" },
+            { LLM_TENSOR_ALTUP_ROUTER,         "blk.%d.altup_router" },
+            { LLM_TENSOR_ALTUP_ROUTER_NORM,    "blk.%d.altup_router_norm" },
+            { LLM_TENSOR_LAUREL_L,             "blk.%d.laurel_l" },
+            { LLM_TENSOR_LAUREL_R,             "blk.%d.laurel_r" },
+            { LLM_TENSOR_LAUREL_POST_NORM,     "blk.%d.laurel_post_norm" },
+        },
+    },
     {
         LLM_ARCH_STARCODER2,
         {
@@ -1553,6 +1631,51 @@ static const std::map<llm_arch, std::map<llm_tensor, const char *>> LLM_TENSOR_N
             { LLM_TENSOR_FFN_UP_SHEXP,       "blk.%d.ffn_up_shexp" },
         },
     },
+    {
+        LLM_ARCH_DOTS1,
+        {
+            { LLM_TENSOR_TOKEN_EMBD,         "token_embd" },
+            { LLM_TENSOR_OUTPUT_NORM,        "output_norm" },
+            { LLM_TENSOR_OUTPUT,             "output" },
+            { LLM_TENSOR_ATTN_NORM,          "blk.%d.attn_norm" },
+            { LLM_TENSOR_ATTN_Q,             "blk.%d.attn_q" },
+            { LLM_TENSOR_ATTN_Q_NORM,        "blk.%d.attn_q_norm" },
+            { LLM_TENSOR_ATTN_K,             "blk.%d.attn_k" },
+            { LLM_TENSOR_ATTN_K_NORM,        "blk.%d.attn_k_norm" },
+            { LLM_TENSOR_ATTN_V,             "blk.%d.attn_v" },
+            { LLM_TENSOR_ATTN_OUT,           "blk.%d.attn_output" },
+            { LLM_TENSOR_FFN_NORM,           "blk.%d.ffn_norm" },
+            { LLM_TENSOR_FFN_GATE,           "blk.%d.ffn_gate" },
+            { LLM_TENSOR_FFN_UP,             "blk.%d.ffn_up" },
+            { LLM_TENSOR_FFN_DOWN,           "blk.%d.ffn_down" },
+            { LLM_TENSOR_FFN_GATE_INP,       "blk.%d.ffn_gate_inp" },
+            { LLM_TENSOR_FFN_GATE_EXPS,      "blk.%d.ffn_gate_exps" },
+            { LLM_TENSOR_FFN_DOWN_EXPS,      "blk.%d.ffn_down_exps" },
+            { LLM_TENSOR_FFN_UP_EXPS,        "blk.%d.ffn_up_exps" },
+            { LLM_TENSOR_FFN_GATE_INP_SHEXP, "blk.%d.ffn_gate_inp_shexp" },
+            { LLM_TENSOR_FFN_GATE_SHEXP,     "blk.%d.ffn_gate_shexp" },
+            { LLM_TENSOR_FFN_DOWN_SHEXP,     "blk.%d.ffn_down_shexp" },
+            { LLM_TENSOR_FFN_UP_SHEXP,       "blk.%d.ffn_up_shexp" },
+            { LLM_TENSOR_FFN_EXP_PROBS_B,    "blk.%d.exp_probs_b" },
+        }
+    },
+    {
+        LLM_ARCH_ERNIE4_5,
+        {
+            { LLM_TENSOR_TOKEN_EMBD,         "token_embd" },
+            { LLM_TENSOR_OUTPUT_NORM,        "output_norm" },
+            { LLM_TENSOR_OUTPUT,             "output" },
+            { LLM_TENSOR_ATTN_NORM,          "blk.%d.attn_norm" },
+            { LLM_TENSOR_ATTN_Q,             "blk.%d.attn_q" },
+            { LLM_TENSOR_ATTN_K,             "blk.%d.attn_k" },
+            { LLM_TENSOR_ATTN_V,             "blk.%d.attn_v" },
+            { LLM_TENSOR_ATTN_OUT,           "blk.%d.attn_output" },
+            { LLM_TENSOR_FFN_NORM,           "blk.%d.ffn_norm" },
+            { LLM_TENSOR_FFN_GATE,           "blk.%d.ffn_gate" },
+            { LLM_TENSOR_FFN_DOWN,           "blk.%d.ffn_down" },
+            { LLM_TENSOR_FFN_UP,             "blk.%d.ffn_up" },
+        },
+    },
     {
         LLM_ARCH_UNKNOWN,
         {
@@ -1681,6 +1804,23 @@ static const std::map<llm_tensor, llm_tensor_info> LLM_TENSOR_INFOS = {
     {LLM_TENSOR_FFN_GATE_EXPS,              {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT_ID}},
     {LLM_TENSOR_FFN_UP_EXPS,                {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT_ID}},
     {LLM_TENSOR_FFN_EXP_PROBS_B,            {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ADD}},
+    // altup / laurel (gemma 3n)
+    {LLM_TENSOR_PER_LAYER_TOKEN_EMBD,       {LLM_TENSOR_LAYER_OUTPUT,    GGML_OP_GET_ROWS}},
+    {LLM_TENSOR_PER_LAYER_MODEL_PROJ,       {LLM_TENSOR_LAYER_OUTPUT,    GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_PER_LAYER_PROJ_NORM,        {LLM_TENSOR_LAYER_OUTPUT,    GGML_OP_MUL}},
+    {LLM_TENSOR_ALTUP_PROJ,                 {LLM_TENSOR_LAYER_OUTPUT,    GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_ALTUP_UNEMBD_PROJ,          {LLM_TENSOR_LAYER_OUTPUT,    GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_PER_LAYER_INP_GATE,         {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_PER_LAYER_PROJ,             {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_PER_LAYER_POST_NORM,        {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_ALTUP_CORRECT_COEF,         {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_ALTUP_CORRECT_SCALE,        {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_ALTUP_PREDICT_COEF,         {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_ALTUP_ROUTER,               {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_ALTUP_ROUTER_NORM,          {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
+    {LLM_TENSOR_LAUREL_L,                   {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_LAUREL_R,                   {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
+    {LLM_TENSOR_LAUREL_POST_NORM,           {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
     // this tensor is loaded for T5, but never used
     {LLM_TENSOR_DEC_CROSS_ATTN_REL_B,       {LLM_TENSOR_LAYER_REPEATING, GGML_OP_NONE}},
     {LLM_TENSOR_CONV1D,                     {LLM_TENSOR_LAYER_INPUT,     GGML_OP_IM2COL}},
@@ -1704,8 +1844,14 @@ static const std::map<llm_tensor, llm_tensor_info> LLM_TENSOR_INFOS = {
 LLM_KV::LLM_KV(llm_arch arch, const char * suffix) : arch(arch), suffix(suffix) {}
 
 std::string LLM_KV::operator()(llm_kv kv) const {
-    return suffix ? ::format(LLM_KV_NAMES.at(kv), LLM_ARCH_NAMES.at(arch), suffix)
-        : ::format(LLM_KV_NAMES.at(kv), LLM_ARCH_NAMES.at(arch));
+    std::string name = ::format(LLM_KV_NAMES.at(kv), LLM_ARCH_NAMES.at(arch));
+
+    if (suffix != nullptr) {
+        name += ".";
+        name += suffix;
+    }
+
+    return name;
 }
 
 std::string LLM_TN_IMPL::str() const {
@@ -1744,3 +1890,25 @@ llm_arch llm_arch_from_string(const std::string & name) {
 const llm_tensor_info & llm_tensor_info_for(llm_tensor tensor) {
     return LLM_TENSOR_INFOS.at(tensor);
 }
+
+bool llm_arch_is_recurrent(const llm_arch & arch) {
+    switch (arch) {
+        case LLM_ARCH_MAMBA:
+        case LLM_ARCH_RWKV6:
+        case LLM_ARCH_RWKV6QWEN2:
+        case LLM_ARCH_RWKV7:
+        case LLM_ARCH_ARWKV7:
+            return true;
+        default:
+            return false;
+    }
+}
+
+bool llm_arch_is_hybrid(const llm_arch & arch) {
+    // TODO: There are currently no hybrid models! Once there are, this will be
+    //  the place to identify them
+    switch (arch) {
+        default:
+            return false;
+    }
+}