fix: use hardware concurrency for automatic CPU threads

README.md

@@ -40,8 +40,8 @@ GGUF models are available at [Audio Embedding GGUF](https://huggingface.co/colle
 Load a converted GGUF and generate real native text or audio embeddings:
 
 ```sh
-clap embedding --model larger_clap_general-q4_k_m.gguf --threads 4 --text 'Sound of a dog'
-clap embedding --model larger_clap_general-q4_k_m.gguf --threads 4 --audio dog.wav
+clap embedding --model larger_clap_general-q4_k_m.gguf --text 'Sound of a dog'
+clap embedding --model larger_clap_general-q4_k_m.gguf --audio dog.wav
 ```
 
 ### Embeddings server
@@ -50,7 +50,7 @@ Start the server with a GGUF supporting the requested modality, such as CLAP,
 CLaMP3 or MuQ-MuLan (default listen address: `127.0.0.1:8080`):
 
 ```sh
-clap serve --model larger_clap_general-q4_k_m.gguf --threads 4 --host 127.0.0.1 --port 8080
+clap serve --model larger_clap_general-q4_k_m.gguf --host 127.0.0.1 --port 8080
 ```
 
 The model is loaded once at startup. Its API identifier is the GGUF **filename**

node/README.md

@@ -17,7 +17,7 @@ npm install clap-cpp
 ```js
 const { CLAP } = require('clap-cpp');
 
-const model = new CLAP('larger_clap_general-q4_k_m.gguf', 4);
+const model = new CLAP('larger_clap_general-q4_k_m.gguf');
 try {
   const textEmbedding = model.encodeText('Sound of a dog');
   const audioEmbedding = model.encodeAudio('dog.wav');

node/test/clap.test.js

@@ -41,7 +41,7 @@ test('CLAP', async (t) => {
     await pipeline(file.stream(), createWriteStream(modelPath));
   }
 
-  const clap = new CLAP(modelPath, 1);
+  const clap = new CLAP(modelPath);
   t.after(() => clap.close());
   const audioEmbedding = clap.encodeAudio(path.join(__dirname, '..', '..', 'assets', 'dog.wav'), 0);
   const textEmbedding1 = clap.encodeText('Sound of a dog');

python/README.md

@@ -26,7 +26,7 @@ pip install "clap-cpp[vulkan]" # Vulkan
 ```python
 from clap_cpp import CLAP
 
-with CLAP("larger_clap_general-q4_k_m.gguf", threads=4) as model:
+with CLAP("larger_clap_general-q4_k_m.gguf") as model:
     text_embedding = model.encode_text("Sound of a dog")
     audio_embedding = model.encode_audio("dog.wav")
 ```

python/tests/test_clap.py

@@ -33,7 +33,7 @@ def test_clap():
     )
 
     audio_path = Path(__file__).resolve().parents[2] / "assets" / "dog.wav"
-    with CLAP(model_path, threads=1) as clap:
+    with CLAP(model_path) as clap:
         audio_embedding = clap.encode_audio(audio_path, 0)
         text_embedding_1 = clap.encode_text("Sound of a dog")
         text_embedding_2 = clap.encode_text("Sound of vaccum cleaner")

src/clap.cpp

@@ -22,7 +22,7 @@ int main(int argc, char **argv) {
     auto *devices = app.add_subcommand("devices", "List available GGML devices");
     auto *serve = app.add_subcommand("serve", "Start the embeddings HTTP server");
     serve->add_option("-m,--model", server_model, "Converted model GGUF")->required()->check(CLI::ExistingFile);
-    serve->add_option("-t,--threads", server_threads, "ggml CPU threads (0: automatic)")
+    serve->add_option("-t,--threads", server_threads, "ggml CPU threads (0: hardware concurrency)")
         ->check(CLI::NonNegativeNumber)
         ->capture_default_str();
     serve->add_option("--device", server_device, "GGML device name, auto or cpu (default: CLAP_DEVICE or auto)");
@@ -35,7 +35,7 @@ int main(int argc, char **argv) {
     embedding->add_option("-m,--model", options.model_path, "Converted CLAP/CLaMP3/MERT/MERT2/MuQ/MuLan GGUF")
         ->required()
         ->check(CLI::ExistingFile);
-    embedding->add_option("-t,--threads", options.threads, "ggml CPU threads (0: automatic)")
+    embedding->add_option("-t,--threads", options.threads, "ggml CPU threads (0: hardware concurrency)")
         ->check(CLI::NonNegativeNumber)
         ->capture_default_str();
     embedding->add_option("--device", options.device, "GGML device name, auto or cpu (default: CLAP_DEVICE or auto)");

src/graph.cpp

@@ -79,8 +79,8 @@ struct GraphRuntime::State {
         cpu.reset(ggml_backend_cpu_init());
         if (!cpu)
             throw std::runtime_error("CLAP: CPU backend initialization failed");
-        ggml_backend_cpu_set_n_threads(
-            cpu.get(), threads > 0 ? threads : int(std::max(1u, std::min(8u, std::thread::hardware_concurrency()))));
+        ggml_backend_cpu_set_n_threads(cpu.get(),
+                                       threads > 0 ? threads : int(std::max(1u, std::thread::hardware_concurrency())));
         if (requested == "auto") {
             for (size_t i = 0; i < ggml_backend_dev_count() && !gpu; ++i) {
                 auto device = ggml_backend_dev_get(i);