whisper : add context param to disable gpu (#1293)

* whisper : check state->ctx_metal not null

* whisper : add whisper_context_params { use_gpu }

* whisper : new API with params & deprecate old API

* examples : use no-gpu param && whisper_init_from_file_with_params

* whisper.objc : enable metal & disable on simulator

* whisper.swiftui, metal : enable metal & support load default.metallib

* whisper.android : use new API

* bindings : use new API

* addon.node : fix build & test

* bindings : updata java binding

* bindings : add missing whisper_context_default_params_by_ref WHISPER_API for java

* metal : use SWIFTPM_MODULE_BUNDLE for GGML_SWIFT and reuse library load

* metal : move bundle var into block

* metal : use SWIFT_PACKAGE instead of GGML_SWIFT

* style : minor updates

---------

Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>

bindings/java/src/main/java/io/github/ggerganov/whispercpp/WhisperContext.java

@@ -4,6 +4,7 @@ import com.sun.jna.Structure;
 import com.sun.jna.ptr.PointerByReference;
 import io.github.ggerganov.whispercpp.ggml.GgmlType;
 import io.github.ggerganov.whispercpp.WhisperModel;
+import io.github.ggerganov.whispercpp.params.WhisperContextParams;
 
 import java.util.List;
 
@@ -23,8 +24,9 @@ public class WhisperContext extends Structure {
     public PointerByReference vocab;
     public PointerByReference state;
 
-    /** populated by whisper_init_from_file() */
+    /** populated by whisper_init_from_file_with_params() */
     String path_model;
+    WhisperContextParams params;
 
 //    public static class ByReference extends WhisperContext implements Structure.ByReference {
 //    }

bindings/java/src/main/java/io/github/ggerganov/whispercpp/WhisperCpp.java

@@ -2,6 +2,7 @@ package io.github.ggerganov.whispercpp;
 
 import com.sun.jna.Native;
 import com.sun.jna.Pointer;
+import io.github.ggerganov.whispercpp.params.WhisperContextParams;
 import io.github.ggerganov.whispercpp.params.WhisperFullParams;
 import io.github.ggerganov.whispercpp.params.WhisperSamplingStrategy;
 
@@ -15,8 +16,9 @@ import java.io.IOException;
 public class WhisperCpp implements AutoCloseable {
     private WhisperCppJnaLibrary lib = WhisperCppJnaLibrary.instance;
     private Pointer ctx = null;
-    private Pointer greedyPointer = null;
-    private Pointer beamPointer = null;
+    private Pointer paramsPointer = null;
+    private Pointer greedyParamsPointer = null;
+    private Pointer beamParamsPointer = null;
 
     public File modelDir() {
         String modelDirPath = System.getenv("XDG_CACHE_HOME");
@@ -31,6 +33,18 @@ public class WhisperCpp implements AutoCloseable {
      * @param modelPath - absolute path, or just the name (eg: "base", "base-en" or "base.en")
      */
     public void initContext(String modelPath) throws FileNotFoundException {
+        initContextImpl(modelPath, getContextDefaultParams());
+    }
+
+    /**
+     * @param modelPath - absolute path, or just the name (eg: "base", "base-en" or "base.en")
+     * @param params - params to use when initialising the context
+     */
+    public void initContext(String modelPath, WhisperContextParams params) throws FileNotFoundException {
+        initContextImpl(modelPath, params);
+    }
+
+    private void initContextImpl(String modelPath, WhisperContextParams params) throws FileNotFoundException {
         if (ctx != null) {
             lib.whisper_free(ctx);
         }
@@ -43,13 +57,26 @@ public class WhisperCpp implements AutoCloseable {
             modelPath = new File(modelDir(), modelPath).getAbsolutePath();
         }
 
-        ctx = lib.whisper_init_from_file(modelPath);
+        ctx = lib.whisper_init_from_file_with_params(modelPath, params);
 
         if (ctx == null) {
             throw new FileNotFoundException(modelPath);
         }
     }
 
+    /**
+     * Provides default params which can be used with `whisper_init_from_file_with_params()` etc.
+     * Because this function allocates memory for the params, the caller must call either:
+     * - call `whisper_free_context_params()`
+     * - `Native.free(Pointer.nativeValue(pointer));`
+     */
+    public WhisperContextParams getContextDefaultParams() {
+        paramsPointer = lib.whisper_context_default_params_by_ref();
+        WhisperContextParams params = new WhisperContextParams(paramsPointer);
+        params.read();
+        return params;
+    }
+    
     /**
      * Provides default params which can be used with `whisper_full()` etc.
      * Because this function allocates memory for the params, the caller must call either:
@@ -63,15 +90,15 @@ public class WhisperCpp implements AutoCloseable {
 
         // whisper_full_default_params_by_ref allocates memory which we need to delete, so only create max 1 pointer for each strategy.
         if (strategy == WhisperSamplingStrategy.WHISPER_SAMPLING_GREEDY) {
-            if (greedyPointer == null) {
-                greedyPointer = lib.whisper_full_default_params_by_ref(strategy.ordinal());
+            if (greedyParamsPointer == null) {
+                greedyParamsPointer = lib.whisper_full_default_params_by_ref(strategy.ordinal());
             }
-            pointer = greedyPointer;
+            pointer = greedyParamsPointer;
         } else {
-            if (beamPointer == null) {
-                beamPointer = lib.whisper_full_default_params_by_ref(strategy.ordinal());
+            if (beamParamsPointer == null) {
+                beamParamsPointer = lib.whisper_full_default_params_by_ref(strategy.ordinal());
             }
-            pointer = beamPointer;
+            pointer = beamParamsPointer;
         }
 
         WhisperFullParams params = new WhisperFullParams(pointer);
@@ -93,13 +120,17 @@ public class WhisperCpp implements AutoCloseable {
     }
 
     private void freeParams() {
-        if (greedyPointer != null) {
-            Native.free(Pointer.nativeValue(greedyPointer));
-            greedyPointer = null;
+        if (paramsPointer != null) {
+            Native.free(Pointer.nativeValue(paramsPointer));
+            paramsPointer = null;
         }
-        if (beamPointer != null) {
-            Native.free(Pointer.nativeValue(beamPointer));
-            beamPointer = null;
+        if (greedyParamsPointer != null) {
+            Native.free(Pointer.nativeValue(greedyParamsPointer));
+            greedyParamsPointer = null;
+        }
+        if (beamParamsPointer != null) {
+            Native.free(Pointer.nativeValue(beamParamsPointer));
+            beamParamsPointer = null;
         }
     }
 

bindings/java/src/main/java/io/github/ggerganov/whispercpp/WhisperCppJnaLibrary.java

@@ -5,6 +5,7 @@ import com.sun.jna.Native;
 import com.sun.jna.Pointer;
 import io.github.ggerganov.whispercpp.model.WhisperModelLoader;
 import io.github.ggerganov.whispercpp.model.WhisperTokenData;
+import io.github.ggerganov.whispercpp.params.WhisperContextParams;
 import io.github.ggerganov.whispercpp.params.WhisperFullParams;
 
 public interface WhisperCppJnaLibrary extends Library {
@@ -13,12 +14,31 @@ public interface WhisperCppJnaLibrary extends Library {
     String whisper_print_system_info();
 
     /**
-     * Allocate (almost) all memory needed for the model by loading from a file.
+     * DEPRECATED. Allocate (almost) all memory needed for the model by loading from a file.
      *
      * @param path_model Path to the model file
      * @return Whisper context on success, null on failure
      */
     Pointer whisper_init_from_file(String path_model);
+    
+    /**
+     * Provides default params which can be used with `whisper_init_from_file_with_params()` etc.
+     * Because this function allocates memory for the params, the caller must call either:
+     * - call `whisper_free_context_params()`
+     * - `Native.free(Pointer.nativeValue(pointer));`
+     */
+    Pointer whisper_context_default_params_by_ref();
+
+    void whisper_free_context_params(Pointer params);
+
+    /**
+     * Allocate (almost) all memory needed for the model by loading from a file.
+     *
+     * @param path_model Path to the model file
+     * @param params     Pointer to whisper_context_params
+     * @return Whisper context on success, null on failure
+     */
+    Pointer whisper_init_from_file_with_params(String path_model, WhisperContextParams params);
 
     /**
      * Allocate (almost) all memory needed for the model by loading from a buffer.

bindings/java/src/main/java/io/github/ggerganov/whispercpp/params/WhisperContextParams.java

@@ -0,0 +1,31 @@
+package io.github.ggerganov.whispercpp.params;
+
+import com.sun.jna.*;
+
+import java.util.Arrays;
+import java.util.List;
+
+/**
+ * Parameters for the whisper_init_from_file_with_params() function.
+ * If you change the order or add new parameters, make sure to update the default values in whisper.cpp:
+ * whisper_context_default_params()
+ */
+public class WhisperContextParams extends Structure {
+
+    public WhisperContextParams(Pointer p) {
+        super(p);
+    }
+
+    /** Use GPU for inference Number (default = true) */
+    public CBool use_gpu;
+
+    /** Use GPU for inference Number (default = true) */
+    public void useGpu(boolean enable) {
+        use_gpu = enable ? CBool.TRUE : CBool.FALSE;
+    }
+
+    @Override
+    protected List<String> getFieldOrder() {
+        return Arrays.asList("use_gpu");
+    }
+}