RubyGems - whispercpp - Versions diffs - 1.3.5 → 1.3.6 - Mend

whispercpp 1.3.5 → 1.3.6

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

Files changed (610) hide show

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

@@ -103,6 +103,7 @@ struct whisper_params {
     bool no_timestamps   = false;
     bool use_gpu         = true;
     bool flash_attn      = true;
+    int32_t gpu_device   = 0;
     bool suppress_nst    = false;
     bool no_context      = true;
     bool no_language_probabilities = false;
@@ -179,6 +180,7 @@ void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params & para
     fprintf(stderr, "  -sns,      --suppress-nst      [%-7s] suppress non-speech tokens\n", params.suppress_nst ? "true" : "false");
     fprintf(stderr, "  -nth N,    --no-speech-thold N [%-7.2f] no speech threshold\n",   params.no_speech_thold);
     fprintf(stderr, "  -ng,       --no-gpu            [%-7s] do not use gpu\n", params.use_gpu ? "false" : "true");
+    fprintf(stderr, "  -dev N,    --device N          [%-7d] GPU device ID (default: 0)\n", params.gpu_device);
     fprintf(stderr, "  -fa,       --flash-attn        [%-7s] enable flash attention\n", params.flash_attn ? "true" : "false");
     fprintf(stderr, "  -nfa,      --no-flash-attn     [%-7s] disable flash attention\n", params.flash_attn ? "false" : "true");
     fprintf(stderr, "  -nlp,      --no-language-probabilities [%-7s] exclude language probabilities from verbose_json output\n", params.no_language_probabilities ? "true" : "false");
@@ -198,6 +200,10 @@ void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params & para
 }
 bool whisper_params_parse(int argc, char ** argv, whisper_params & params, server_params & sparams) {
+    if (const char * env_device = std::getenv("WHISPER_ARG_DEVICE")) {
+        params.gpu_device = std::stoi(env_device);
+    }
     for (int i = 1; i < argc; i++) {
         std::string arg = argv[i];
@@ -237,6 +243,7 @@ bool whisper_params_parse(int argc, char ** argv, whisper_params & params, serve
         else if (arg == "-oved" || arg == "--ov-e-device")     { params.openvino_encode_device = argv[++i]; }
         else if (arg == "-dtw"  || arg == "--dtw")             { params.dtw             = argv[++i]; }
         else if (arg == "-ng"   || arg == "--no-gpu")          { params.use_gpu         = false; }
+        else if (arg == "-dev"  || arg == "--device")          { params.gpu_device      = std::stoi(argv[++i]); }
         else if (arg == "-fa"   || arg == "--flash-attn")      { params.flash_attn      = true; }
         else if (arg == "-nfa"  || arg == "--no-flash-attn")   { params.flash_attn      = false; }
         else if (arg == "-sns"  || arg == "--suppress-nst")    { params.suppress_nst    = true; }
@@ -643,6 +650,7 @@ int main(int argc, char ** argv) {
     struct whisper_context_params cparams = whisper_context_default_params();
     cparams.use_gpu    = params.use_gpu;
+    cparams.gpu_device = params.gpu_device;
     cparams.flash_attn = params.flash_attn;
     if (!params.dtw.empty()) {
@@ -740,9 +748,9 @@ int main(int argc, char ** argv) {
     <body>
         <h1>Whisper.cpp Server</h1>
-        <h2>/inference</h2>
+        <h2>)" + sparams.request_path + sparams.inference_path + R"(</h2>
         <pre>
-    curl 127.0.0.1:)" + std::to_string(sparams.port) + R"(/inference \
+    curl 127.0.0.1:)" + std::to_string(sparams.port) + sparams.request_path + sparams.inference_path + R"( \
     -H "Content-Type: multipart/form-data" \
     -F file="@&lt;file-path&gt;" \
     -F temperature="0.0" \
@@ -759,7 +767,7 @@ int main(int argc, char ** argv) {
         <div>
             <h2>Try it out</h2>
-            <form action="/inference" method="POST" enctype="multipart/form-data">
+            <form action=")" + sparams.request_path + sparams.inference_path + R"(" method="POST" enctype="multipart/form-data">
                 <label for="file">Choose an audio file:</label>
                 <input type="file" id="file" name="file" accept="audio/*" required><br>
@@ -803,6 +811,7 @@ int main(int argc, char ** argv) {
         {
             fprintf(stderr, "error: no 'file' field in the request\n");
             const std::string error_resp = "{\"error\":\"no 'file' field in the request\"}";
+            res.status = 400;
             res.set_content(error_resp, "application/json");
             return;
         }
@@ -829,6 +838,7 @@ int main(int argc, char ** argv) {
             std::string error_resp = "{\"error\":\"Failed to execute ffmpeg command.\"}";
             const bool is_converted = convert_to_wav(temp_filename, error_resp);
             if (!is_converted) {
+                res.status = 500;
                 res.set_content(error_resp, "application/json");
                 return;
             }
@@ -838,6 +848,7 @@ int main(int argc, char ** argv) {
             {
                 fprintf(stderr, "error: failed to read WAV file '%s'\n", temp_filename.c_str());
                 const std::string error_resp = "{\"error\":\"failed to read WAV file\"}";
+                res.status = 400;
                 res.set_content(error_resp, "application/json");
                 std::remove(temp_filename.c_str());
                 return;
@@ -849,6 +860,7 @@ int main(int argc, char ** argv) {
             {
                 fprintf(stderr, "error: failed to read audio data\n");
                 const std::string error_resp = "{\"error\":\"failed to read audio data\"}";
+                res.status = 400;
                 res.set_content(error_resp, "application/json");
                 return;
             }
@@ -927,7 +939,7 @@ int main(int argc, char ** argv) {
             wparams.logprob_thold    = params.logprob_thold;
             wparams.no_timestamps    = params.no_timestamps;
-            wparams.token_timestamps = !params.no_timestamps && params.response_format == vjson_format;
+            wparams.token_timestamps = !params.no_timestamps;
             wparams.no_context       = params.no_context;
             wparams.suppress_nst     = params.suppress_nst;
@@ -1119,6 +1131,7 @@ int main(int argc, char ** argv) {
         {
             fprintf(stderr, "error: no 'model' field in the request\n");
             const std::string error_resp = "{\"error\":\"no 'model' field in the request\"}";
+            res.status = 400;
             res.set_content(error_resp, "application/json");
             return;
         }
@@ -1127,6 +1140,7 @@ int main(int argc, char ** argv) {
         {
             fprintf(stderr, "error: 'model': %s not found!\n", model.c_str());
             const std::string error_resp = "{\"error\":\"model not found!\"}";
+            res.status = 400;
             res.set_content(error_resp, "application/json");
             return;
         }

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

@@ -22,18 +22,19 @@ if (WHISPER_SDL2)
         llama-kv-cache-iswa.cpp
         llama-memory-recurrent.cpp
         llama-memory-hybrid.cpp
+        llama-memory-hybrid-iswa.cpp
         llama-memory.cpp
         llama-mmap.cpp
         llama-model-loader.cpp
         llama-model-saver.cpp
         llama-model.cpp
         llama-quant.cpp
-        llama-sampling.cpp
+        llama-sampler.cpp
         llama-vocab.cpp
         unicode.cpp
         unicode-data.cpp
         ${SRC_MODELS})
-    target_include_directories(${TARGET} PRIVATE ${SDL2_INCLUDE_DIRS})
+    target_include_directories(${TARGET} PRIVATE . ${SDL2_INCLUDE_DIRS})
     target_link_libraries(${TARGET} PRIVATE common common-sdl whisper ${SDL2_LIBRARIES} ${CMAKE_THREAD_LIBS_INIT})
     install(TARGETS ${TARGET} RUNTIME)

data/ext/sources/examples/talk-llama/llama-adapter.cpp CHANGED Viewed

@@ -146,11 +146,9 @@ llama_adapter_lora_weight * llama_adapter_lora::get_weight(ggml_tensor * w) {
     return nullptr;
 }
-static void llama_adapter_lora_init_impl(const char * path_lora, llama_adapter_lora & adapter) {
+static void llama_adapter_lora_init_impl(llama_model & model, const char * path_lora, llama_adapter_lora & adapter) {
     LLAMA_LOG_INFO("%s: loading lora adapter from '%s' ...\n", __func__, path_lora);
-    llama_model & model = adapter.model;
     ggml_context * ctx_init;
     gguf_init_params meta_gguf_params = {
         /* .no_alloc = */ true,
@@ -413,17 +411,17 @@ static void llama_adapter_lora_init_impl(const char * path_lora, llama_adapter_l
         }
     }
-    // update number of nodes used
-    model.n_lora_nodes += adapter.get_n_nodes();
+    // register adapter with model
+    model.loras.insert(&adapter);
     LLAMA_LOG_INFO("%s: loaded %zu tensors from lora file\n", __func__, adapter.ab_map.size()*2);
 }
 llama_adapter_lora * llama_adapter_lora_init(llama_model * model, const char * path_lora) {
-    llama_adapter_lora * adapter = new llama_adapter_lora(*model);
+    llama_adapter_lora * adapter = new llama_adapter_lora();
     try {
-        llama_adapter_lora_init_impl(path_lora, *adapter);
+        llama_adapter_lora_init_impl(*model, path_lora, *adapter);
         return adapter;
     } catch (const std::exception & err) {
         LLAMA_LOG_ERROR("%s: failed to apply lora adapter: %s\n", __func__, err.what());
@@ -473,12 +471,8 @@ int32_t llama_adapter_meta_val_str_by_index(const llama_adapter_lora * adapter,
     return snprintf(buf, buf_size, "%s", it->second.c_str());
 }
-void llama_adapter_lora_free(llama_adapter_lora * adapter) {
-    // update number of nodes used
-    GGML_ASSERT(adapter->model.n_lora_nodes >= adapter->get_n_nodes());
-    adapter->model.n_lora_nodes -= adapter->get_n_nodes();
-    delete adapter;
+void llama_adapter_lora_free(llama_adapter_lora *) {
+    // deprecated: adapters are freed by llama_model's destructor
 }
 uint64_t llama_adapter_get_alora_n_invocation_tokens(const struct llama_adapter_lora * adapter) {

data/ext/sources/examples/talk-llama/llama-adapter.h CHANGED Viewed

@@ -39,6 +39,8 @@ private:
     std::vector<ggml_tensor *> tensors; // per layer
 };
+using llama_adapter_cvec_ptr = std::shared_ptr<llama_adapter_cvec>;
 //
 // llama_adapter_lora
 //
@@ -59,8 +61,6 @@ struct llama_adapter_lora_weight {
 };
 struct llama_adapter_lora {
-    llama_model & model;
     // map tensor name to lora_a_b
     std::unordered_map<std::string, llama_adapter_lora_weight> ab_map;
@@ -75,7 +75,7 @@ struct llama_adapter_lora {
     // activated lora (aLoRA)
     std::vector<llama_token> alora_invocation_tokens;
-    llama_adapter_lora(llama_model & model) : model(model) {}
+    llama_adapter_lora() = default;
     ~llama_adapter_lora() = default;
     llama_adapter_lora_weight * get_weight(ggml_tensor * w);
@@ -86,3 +86,4 @@ struct llama_adapter_lora {
 };
 using llama_adapter_loras = std::unordered_map<llama_adapter_lora *, float>;
+using llama_adapter_loras_ptr = std::unique_ptr<llama_adapter_loras>;