release : v1.5.4

Makefile

@@ -42,7 +42,7 @@ publish: publish-trigger
 		\n\
 		cd /path/to/whisper.cpp/bindings/ios\n\
 		git commit\n\
-		git tag 1.5.3\n\
+		git tag 1.5.4\n\
 		git push origin master --tags\n\
 		"
 

Sources/whisper/coreml/whisper-encoder.mm

@@ -24,9 +24,9 @@
 
     // select which device to run the Core ML model on
     MLModelConfiguration *config = [[MLModelConfiguration alloc] init];
-    config.computeUnits = MLComputeUnitsCPUAndGPU;
+    // config.computeUnits = MLComputeUnitsCPUAndGPU;
     //config.computeUnits = MLComputeUnitsCPUAndNeuralEngine;
-    //config.computeUnits = MLComputeUnitsAll;
+    config.computeUnits = MLComputeUnitsAll;
 
     const void * data = CFBridgingRetain([[whisper_encoder_impl alloc] initWithContentsOfURL:url_model configuration:config error:nil]);
 

Sources/whisper/ggml-metal.m

@@ -258,14 +258,14 @@ static void ggml_metal_log(enum ggml_log_level level, const char * format, ...){
         bundle = [NSBundle bundleForClass:[GGMLMetalClass class]];
 #endif
         NSError * error = nil;
-        NSString * libPath = [bundle pathForResource:@"ggml" ofType:@"metallib"];
+        NSString * libPath = [bundle pathForResource:@"default" ofType:@"metallib"];
         if (libPath != nil) {
             // pre-compiled library found
             NSURL * libURL = [NSURL fileURLWithPath:libPath];
             GGML_METAL_LOG_INFO("%s: loading '%s'\n", __func__, [libPath UTF8String]);
             ctx->library = [ctx->device newLibraryWithURL:libURL error:&error];
         } else {
-            GGML_METAL_LOG_INFO("%s: ggml.metallib not found, loading from source\n", __func__);
+            GGML_METAL_LOG_INFO("%s: default.metallib not found, loading from source\n", __func__);
 
             NSString * sourcePath;
             NSString * ggmlMetalPathResources = [[NSProcessInfo processInfo].environment objectForKey:@"GGML_METAL_PATH_RESOURCES"];
@@ -295,7 +295,7 @@ static void ggml_metal_log(enum ggml_log_level level, const char * format, ...){
 #endif
             // try to disable fast-math
             // NOTE: this seems to have no effect whatsoever
-            //       instead, in order to disable fast-math, we have to build ggml.metallib from the command line
+            //       instead, in order to disable fast-math, we have to build default.metallib from the command line
             //       using xcrun -sdk macosx metal -fno-fast-math -c ggml-metal.metal -o ggml-metal.air
             //       and go through the "pre-compiled library found" path above
             //[options setFastMathEnabled:false];

Sources/whisper/ggml-quants.h

@@ -70,7 +70,7 @@ static_assert(sizeof(block_q8_1) == 2*sizeof(float) + QK8_1, "wrong q8_1 block s
 // 2-bit quantization
 // weight is represented as x = a * q + b
 // 16 blocks of 16 elements each
-// Effectively 2.5625 bits per weight
+// Effectively 2.625 bits per weight
 typedef struct {
     uint8_t scales[QK_K/16]; // scales and mins, quantized with 4 bits
     uint8_t qs[QK_K/4];      // quants

Sources/whisper/whisper.cpp

@@ -3821,6 +3821,7 @@ void whisper_reset_timings(struct whisper_context * ctx) {
         ctx->state->t_sample_us = 0;
         ctx->state->t_encode_us = 0;
         ctx->state->t_decode_us = 0;
+        ctx->state->t_batchd_us = 0;
         ctx->state->t_prompt_us = 0;
         ctx->state->n_sample = 0;
         ctx->state->n_encode = 0;