Merge pull request #169 from arizhih/whisper-cpp-latest

feat: Add support for new whisper.cpp project structure

Cargo.toml

@@ -4,7 +4,7 @@ exclude = ["examples/full_usage"]
 
 [package]
 name = "whisper-rs"
-version = "0.12.1"
+version = "0.13.0"
 edition = "2021"
 description = "Rust bindings for whisper.cpp"
 license = "Unlicense"
@@ -14,7 +14,7 @@ repository = "https://github.com/tazz4843/whisper-rs"
 # See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
 
 [dependencies]
-whisper-rs-sys = { path = "sys", version = "0.10.1" }
+whisper-rs-sys = { path = "sys", version = "0.11.0" }
 log = { version = "0.4", optional = true }
 tracing = { version = "0.1", optional = true }
 
@@ -23,19 +23,19 @@ hound = "3.5.0"
 rand = "0.8.4"
 
 [features]
-default = []
+default = ["openmp"]
 
 raw-api = []
 coreml = ["whisper-rs-sys/coreml"]
 cuda = ["whisper-rs-sys/cuda", "_gpu"]
 hipblas = ["whisper-rs-sys/hipblas", "_gpu"]
-opencl = ["whisper-rs-sys/opencl"]
 openblas = ["whisper-rs-sys/openblas"]
 metal = ["whisper-rs-sys/metal", "_gpu"]
 _gpu = []
 test-with-tiny-model = []
 whisper-cpp-log = ["dep:log"]
 whisper-cpp-tracing = ["dep:tracing"]
+openmp = ["whisper-rs-sys/openmp"]
 
 [package.metadata.docs.rs]
 features = ["simd"]

README.md

@@ -72,8 +72,6 @@ All disabled by default unless otherwise specified.
   as whisper-rs-sys may be upgraded to a breaking version in a patch release of whisper-rs.
 * `cuda`: enable CUDA support. Implicitly enables hidden GPU flag at runtime.
 * `hipblas`: enable ROCm/hipBLAS support. Only available on linux. Implicitly enables hidden GPU flag at runtime.
-* `opencl`: enable OpenCL support. Upstream whisper.cpp does not treat OpenCL as a GPU, so it is always enabled at
-  runtime.
 * `openblas`: enable OpenBLAS support.
 * `metal`: enable Metal support. Implicitly enables hidden GPU flag at runtime.
 * `whisper-cpp-log`: allows hooking into whisper.cpp's log output and sending it to the `log` backend. Requires calling

src/standalone.rs

@@ -110,7 +110,6 @@ pub struct SystemInfo {
     pub fma: bool,
     pub f16c: bool,
     pub blas: bool,
-    pub clblast: bool,
     pub cuda: bool,
 }
 
@@ -123,7 +122,6 @@ impl Default for SystemInfo {
                 fma: whisper_rs_sys::ggml_cpu_has_fma() != 0,
                 f16c: whisper_rs_sys::ggml_cpu_has_f16c() != 0,
                 blas: whisper_rs_sys::ggml_cpu_has_blas() != 0,
-                clblast: whisper_rs_sys::ggml_cpu_has_clblast() != 0,
                 cuda: whisper_rs_sys::ggml_cpu_has_cuda() != 0,
             }
         }

src/whisper_ctx.rs

@@ -471,9 +471,6 @@ unsafe impl Sync for WhisperInnerContext {}
 
 pub struct WhisperContextParameters<'a> {
     /// Use GPU if available.
-    ///
-    /// **Warning**: Does not have an effect if OpenCL is selected as GPU backend
-    /// (in that case, GPU is always enabled).
     pub use_gpu: bool,
     /// Enable flash attention, default false
     ///

src/whisper_params.rs

@@ -222,16 +222,6 @@ impl<'a, 'b> FullParams<'a, 'b> {
         self.fp.max_tokens = max_tokens;
     }
 
-    /// # EXPERIMENTAL
-    ///
-    /// Speed up audio ~2x by using phase vocoder.
-    /// Note that this can significantly reduce the accuracy of the transcription.
-    ///
-    /// Defaults to false.
-    pub fn set_speed_up(&mut self, speed_up: bool) {
-        self.fp.speed_up = speed_up;
-    }
-
     /// # EXPERIMENTAL
     ///
     /// Enables debug mode, such as dumping the log mel spectrogram.
@@ -244,7 +234,6 @@ impl<'a, 'b> FullParams<'a, 'b> {
     /// # EXPERIMENTAL
     ///
     /// Overwrite the audio context size. 0 = default.
-    /// As with [set_speed_up](FullParams::set_speed_up), this can significantly reduce the accuracy of the transcription.
     ///
     /// Defaults to 0.
     pub fn set_audio_ctx(&mut self, audio_ctx: c_int) {

src/whisper_state.rs

@@ -64,45 +64,6 @@ impl WhisperState {
         }
     }
 
-    /// Convert raw PCM audio (floating point 32 bit) to log mel spectrogram.
-    /// Applies a Phase Vocoder to speed up the audio x2.
-    /// The resulting spectrogram is stored in the context transparently.
-    ///
-    /// # Arguments
-    /// * pcm: The raw PCM audio.
-    /// * threads: How many threads to use. Defaults to 1. Must be at least 1, returns an error otherwise.
-    ///
-    /// # Returns
-    /// Ok(()) on success, Err(WhisperError) on failure.
-    ///
-    /// # C++ equivalent
-    /// `int whisper_pcm_to_mel(struct whisper_context * ctx, const float * samples, int n_samples, int n_threads)`
-    pub fn pcm_to_mel_phase_vocoder(
-        &mut self,
-        pcm: &[f32],
-        threads: usize,
-    ) -> Result<(), WhisperError> {
-        if threads < 1 {
-            return Err(WhisperError::InvalidThreadCount);
-        }
-        let ret = unsafe {
-            whisper_rs_sys::whisper_pcm_to_mel_phase_vocoder_with_state(
-                self.ctx.ctx,
-                self.ptr,
-                pcm.as_ptr(),
-                pcm.len() as c_int,
-                threads as c_int,
-            )
-        };
-        if ret == -1 {
-            Err(WhisperError::UnableToCalculateSpectrogram)
-        } else if ret == 0 {
-            Ok(())
-        } else {
-            Err(WhisperError::GenericError(ret))
-        }
-    }
-
     /// This can be used to set a custom log mel spectrogram inside the provided whisper state.
     /// Use this instead of whisper_pcm_to_mel() if you want to provide your own log mel spectrogram.
     ///

sys/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "whisper-rs-sys"
-version = "0.10.1"
+version = "0.11.0"
 edition = "2021"
 description = "Rust bindings for whisper.cpp (FFI bindings)"
 license = "Unlicense"
@@ -10,29 +10,28 @@ links = "whisper"
 include = [
     "whisper.cpp/bindings/javascript/package-tmpl.json",
     "whisper.cpp/bindings/CMakeLists.txt",
-    "whisper.cpp/cmake",
-    "whisper.cpp/coreml",
     "whisper.cpp/CMakeLists.txt",
-    "whisper.cpp/ggml.c",
+    "whisper.cpp/cmake",
-    "whisper.cpp/ggml.h",
+    "whisper.cpp/src/coreml",
-    "whisper.cpp/ggml-alloc.c",
+    "whisper.cpp/src/CMakeLists.txt",
-    "whisper.cpp/ggml-alloc.h",
+    "whisper.cpp/src/whisper.cpp",
-    "whisper.cpp/ggml-backend.c",
+    "whisper.cpp/include/whisper.h",
-    "whisper.cpp/ggml-backend.h",
+    "whisper.cpp/ggml/src/ggml.c",
-    "whisper.cpp/ggml-backend-impl.h",
+    "whisper.cpp/ggml/src/ggml-alloc.c",
-    "whisper.cpp/ggml-cuda.cu",
+    "whisper.cpp/ggml/src/ggml-backend.c",
-    "whisper.cpp/ggml-cuda.h",
+    "whisper.cpp/ggml/src/ggml-cuda.cu",
-    "whisper.cpp/ggml-impl.h",
+    "whisper.cpp/ggml/src/ggml-impl.h",
-    "whisper.cpp/ggml-metal.h",
+    "whisper.cpp/ggml/src/ggml-metal.m",
-    "whisper.cpp/ggml-metal.m",
+    "whisper.cpp/ggml/src/ggml-metal.metal",
-    "whisper.cpp/ggml-metal.metal",
+    "whisper.cpp/ggml/src/ggml-quants.h",
-    "whisper.cpp/ggml-opencl.cpp",
+    "whisper.cpp/ggml/src/ggml-quants.c",
-    "whisper.cpp/ggml-opencl.h",
+    "whisper.cpp/ggml/include/ggml.h",
-    "whisper.cpp/ggml-quants.h",
+    "whisper.cpp/ggml/include/ggml-alloc.h",
-    "whisper.cpp/ggml-quants.c",
+    "whisper.cpp/ggml/include/ggml-backend.h",
+    "whisper.cpp/ggml/include/ggml-backend-impl.h",
+    "whisper.cpp/ggml/include/ggml-cuda.h",
+    "whisper.cpp/ggml/include/ggml-metal.h",
     "whisper.cpp/LICENSE",
-    "whisper.cpp/whisper.cpp",
-    "whisper.cpp/whisper.h",
     "src/*.rs",
     "build.rs",
     "wrapper.h",
@@ -44,10 +43,10 @@ include = [
 coreml = []
 cuda = []
 hipblas = []
-opencl = []
 openblas = []
 metal = []
 force-debug = []
+openmp = []
 
 [build-dependencies]
 cmake = "0.1"

sys/build.rs

@@ -30,11 +30,6 @@ fn main() {
 
     #[cfg(feature = "coreml")]
     println!("cargo:rustc-link-lib=static=whisper.coreml");
-    #[cfg(feature = "opencl")]
-    {
-        println!("cargo:rustc-link-lib=clblast");
-        println!("cargo:rustc-link-lib=OpenCL");
-    }
     #[cfg(feature = "openblas")]
     {
         if let Ok(openblas_path) = env::var("OPENBLAS_PATH") {
@@ -91,6 +86,13 @@ fn main() {
         }
     }
 
+    #[cfg(feature = "openmp")]
+    {
+        if target.contains("gnu") {
+            println!("cargo:rustc-link-lib=gomp");
+        }
+    }
+
     println!("cargo:rerun-if-changed=wrapper.h");
 
     let out = PathBuf::from(env::var("OUT_DIR").unwrap());
@@ -114,10 +116,11 @@ fn main() {
         let bindings = bindgen::Builder::default().header("wrapper.h");
 
         #[cfg(feature = "metal")]
-        let bindings = bindings.header("whisper.cpp/ggml-metal.h");
+        let bindings = bindings.header("whisper.cpp/ggml/include/ggml-metal.h");
 
         let bindings = bindings
-            .clang_arg("-I./whisper.cpp")
+            .clang_arg("-I./whisper.cpp/include")
+            .clang_arg("-I./whisper.cpp/ggml/include")
             .parse_callbacks(Box::new(bindgen::CargoCallbacks::new()))
             .generate();
 
@@ -160,11 +163,11 @@ fn main() {
     }
 
     if cfg!(feature = "cuda") {
-        config.define("WHISPER_CUDA", "ON");
+        config.define("GGML_CUDA", "ON");
     }
 
     if cfg!(feature = "hipblas") {
-        config.define("WHISPER_HIPBLAS", "ON");
+        config.define("GGML_HIPBLAS", "ON");
         config.define("CMAKE_C_COMPILER", "hipcc");
         config.define("CMAKE_CXX_COMPILER", "hipcc");
         println!("cargo:rerun-if-env-changed=AMDGPU_TARGETS");
@@ -174,20 +177,16 @@ fn main() {
     }
 
     if cfg!(feature = "openblas") {
-        config.define("WHISPER_OPENBLAS", "ON");
+        config.define("GGML_BLAS", "ON");
-    }
-
-    if cfg!(feature = "opencl") {
-        config.define("WHISPER_CLBLAST", "ON");
     }
 
     if cfg!(feature = "metal") {
-        config.define("WHISPER_METAL", "ON");
+        config.define("GGML_METAL", "ON");
-        config.define("WHISPER_METAL_NDEBUG", "ON");
+        config.define("GGML_METAL_NDEBUG", "ON");
-        config.define("WHISPER_METAL_EMBED_LIBRARY", "ON");
+        config.define("GGML_METAL_EMBED_LIBRARY", "ON");
     } else {
         // Metal is enabled by default, so we need to explicitly disable it
-        config.define("WHISPER_METAL", "OFF");
+        config.define("GGML_METAL", "OFF");
     }
 
     if cfg!(debug_assertions) || cfg!(feature = "force-debug") {
@@ -207,18 +206,17 @@ fn main() {
         }
     }
 
+    if cfg!(not(feature = "openmp")) {
+        config.define("GGML_OPENMP", "OFF");
+    }
+
     let destination = config.build();
 
-    if target.contains("window") && !target.contains("gnu") {
+    add_link_search_path(&out.join("lib")).unwrap();
-        println!(
+
-            "cargo:rustc-link-search={}",
-            out.join("build").join("Release").display()
-        );
-    } else {
-        println!("cargo:rustc-link-search={}", out.join("build").display());
-    }
     println!("cargo:rustc-link-search=native={}", destination.display());
     println!("cargo:rustc-link-lib=static=whisper");
+    println!("cargo:rustc-link-lib=static=ggml");
 
     // for whatever reason this file is generated during build and triggers cargo complaining
     _ = std::fs::remove_file("bindings/javascript/package.json");
@@ -236,3 +234,13 @@ fn get_cpp_link_stdlib(target: &str) -> Option<&'static str> {
         Some("stdc++")
     }
 }
+
+fn add_link_search_path(dir: &std::path::Path) -> std::io::Result<()> {
+    if dir.is_dir() {
+        println!("cargo:rustc-link-search={}", dir.display());
+        for entry in std::fs::read_dir(dir)? {
+            add_link_search_path(&entry?.path())?;
+        }
+    }
+    Ok(())
+}

sys/src/bindings.rs

@@ -1,4 +1,4 @@
-/* automatically generated by rust-bindgen 0.69.1 */
+/* automatically generated by rust-bindgen 0.69.4 */
 
 pub const __bool_true_false_are_defined: u32 = 1;
 pub const true_: u32 = 1;
@@ -161,8 +161,10 @@ pub const GGML_KQ_MASK_PAD: u32 = 32;
 pub const GGML_N_TASKS_MAX: i32 = -1;
 pub const WHISPER_SAMPLE_RATE: u32 = 16000;
 pub const WHISPER_N_FFT: u32 = 400;
+pub const WHISPER_N_FFT_HALF: u32 = 201;
 pub const WHISPER_HOP_LENGTH: u32 = 160;
 pub const WHISPER_CHUNK_SIZE: u32 = 30;
+pub const WHISPER_N_SAMPLES: u32 = 480000;
 pub type wchar_t = ::std::os::raw::c_int;
 #[repr(C)]
 #[repr(align(16))]
@@ -1289,6 +1291,14 @@ extern "C" {
 extern "C" {
     pub fn __overflow(arg1: *mut FILE, arg2: ::std::os::raw::c_int) -> ::std::os::raw::c_int;
 }
+extern "C" {
+    pub fn ggml_abort(
+        file: *const ::std::os::raw::c_char,
+        line: ::std::os::raw::c_int,
+        fmt: *const ::std::os::raw::c_char,
+        ...
+    );
+}
 pub const ggml_status_GGML_STATUS_ALLOC_FAILED: ggml_status = -2;
 pub const ggml_status_GGML_STATUS_FAILED: ggml_status = -1;
 pub const ggml_status_GGML_STATUS_SUCCESS: ggml_status = 0;
@@ -1349,6 +1359,9 @@ extern "C" {
 extern "C" {
     pub fn ggml_bf16_to_fp32_row(arg1: *const ggml_bf16_t, arg2: *mut f32, arg3: i64);
 }
+extern "C" {
+    pub fn ggml_fp32_to_bf16_row_ref(arg1: *const f32, arg2: *mut ggml_bf16_t, arg3: i64);
+}
 extern "C" {
     pub fn ggml_fp32_to_bf16_row(arg1: *const f32, arg2: *mut ggml_bf16_t, arg3: i64);
 }
@@ -1386,7 +1399,10 @@ pub const ggml_type_GGML_TYPE_I64: ggml_type = 27;
 pub const ggml_type_GGML_TYPE_F64: ggml_type = 28;
 pub const ggml_type_GGML_TYPE_IQ1_M: ggml_type = 29;
 pub const ggml_type_GGML_TYPE_BF16: ggml_type = 30;
-pub const ggml_type_GGML_TYPE_COUNT: ggml_type = 31;
+pub const ggml_type_GGML_TYPE_Q4_0_4_4: ggml_type = 31;
+pub const ggml_type_GGML_TYPE_Q4_0_4_8: ggml_type = 32;
+pub const ggml_type_GGML_TYPE_Q4_0_8_8: ggml_type = 33;
+pub const ggml_type_GGML_TYPE_COUNT: ggml_type = 34;
 pub type ggml_type = ::std::os::raw::c_uint;
 pub const ggml_prec_GGML_PREC_DEFAULT: ggml_prec = 0;
 pub const ggml_prec_GGML_PREC_F32: ggml_prec = 1;
@@ -1419,6 +1435,9 @@ pub const ggml_ftype_GGML_FTYPE_MOSTLY_IQ2_S: ggml_ftype = 21;
 pub const ggml_ftype_GGML_FTYPE_MOSTLY_IQ4_XS: ggml_ftype = 22;
 pub const ggml_ftype_GGML_FTYPE_MOSTLY_IQ1_M: ggml_ftype = 23;
 pub const ggml_ftype_GGML_FTYPE_MOSTLY_BF16: ggml_ftype = 24;
+pub const ggml_ftype_GGML_FTYPE_MOSTLY_Q4_0_4_4: ggml_ftype = 25;
+pub const ggml_ftype_GGML_FTYPE_MOSTLY_Q4_0_4_8: ggml_ftype = 26;
+pub const ggml_ftype_GGML_FTYPE_MOSTLY_Q4_0_8_8: ggml_ftype = 27;
 pub type ggml_ftype = ::std::os::raw::c_int;
 pub const ggml_op_GGML_OP_NONE: ggml_op = 0;
 pub const ggml_op_GGML_OP_DUP: ggml_op = 1;
@@ -1475,28 +1494,26 @@ pub const ggml_op_GGML_OP_ARANGE: ggml_op = 51;
 pub const ggml_op_GGML_OP_TIMESTEP_EMBEDDING: ggml_op = 52;
 pub const ggml_op_GGML_OP_ARGSORT: ggml_op = 53;
 pub const ggml_op_GGML_OP_LEAKY_RELU: ggml_op = 54;
-pub const ggml_op_GGML_OP_FLASH_ATTN: ggml_op = 55;
+pub const ggml_op_GGML_OP_FLASH_ATTN_EXT: ggml_op = 55;
-pub const ggml_op_GGML_OP_FLASH_ATTN_EXT: ggml_op = 56;
+pub const ggml_op_GGML_OP_FLASH_ATTN_BACK: ggml_op = 56;
-pub const ggml_op_GGML_OP_FLASH_FF: ggml_op = 57;
+pub const ggml_op_GGML_OP_SSM_CONV: ggml_op = 57;
-pub const ggml_op_GGML_OP_FLASH_ATTN_BACK: ggml_op = 58;
+pub const ggml_op_GGML_OP_SSM_SCAN: ggml_op = 58;
-pub const ggml_op_GGML_OP_SSM_CONV: ggml_op = 59;
+pub const ggml_op_GGML_OP_WIN_PART: ggml_op = 59;
-pub const ggml_op_GGML_OP_SSM_SCAN: ggml_op = 60;
+pub const ggml_op_GGML_OP_WIN_UNPART: ggml_op = 60;
-pub const ggml_op_GGML_OP_WIN_PART: ggml_op = 61;
+pub const ggml_op_GGML_OP_GET_REL_POS: ggml_op = 61;
-pub const ggml_op_GGML_OP_WIN_UNPART: ggml_op = 62;
+pub const ggml_op_GGML_OP_ADD_REL_POS: ggml_op = 62;
-pub const ggml_op_GGML_OP_GET_REL_POS: ggml_op = 63;
+pub const ggml_op_GGML_OP_UNARY: ggml_op = 63;
-pub const ggml_op_GGML_OP_ADD_REL_POS: ggml_op = 64;
+pub const ggml_op_GGML_OP_MAP_UNARY: ggml_op = 64;
-pub const ggml_op_GGML_OP_UNARY: ggml_op = 65;
+pub const ggml_op_GGML_OP_MAP_BINARY: ggml_op = 65;
-pub const ggml_op_GGML_OP_MAP_UNARY: ggml_op = 66;
+pub const ggml_op_GGML_OP_MAP_CUSTOM1_F32: ggml_op = 66;
-pub const ggml_op_GGML_OP_MAP_BINARY: ggml_op = 67;
+pub const ggml_op_GGML_OP_MAP_CUSTOM2_F32: ggml_op = 67;
-pub const ggml_op_GGML_OP_MAP_CUSTOM1_F32: ggml_op = 68;
+pub const ggml_op_GGML_OP_MAP_CUSTOM3_F32: ggml_op = 68;
-pub const ggml_op_GGML_OP_MAP_CUSTOM2_F32: ggml_op = 69;
+pub const ggml_op_GGML_OP_MAP_CUSTOM1: ggml_op = 69;
-pub const ggml_op_GGML_OP_MAP_CUSTOM3_F32: ggml_op = 70;
+pub const ggml_op_GGML_OP_MAP_CUSTOM2: ggml_op = 70;
-pub const ggml_op_GGML_OP_MAP_CUSTOM1: ggml_op = 71;
+pub const ggml_op_GGML_OP_MAP_CUSTOM3: ggml_op = 71;
-pub const ggml_op_GGML_OP_MAP_CUSTOM2: ggml_op = 72;
+pub const ggml_op_GGML_OP_CROSS_ENTROPY_LOSS: ggml_op = 72;
-pub const ggml_op_GGML_OP_MAP_CUSTOM3: ggml_op = 73;
+pub const ggml_op_GGML_OP_CROSS_ENTROPY_LOSS_BACK: ggml_op = 73;
-pub const ggml_op_GGML_OP_CROSS_ENTROPY_LOSS: ggml_op = 74;
+pub const ggml_op_GGML_OP_COUNT: ggml_op = 74;
-pub const ggml_op_GGML_OP_CROSS_ENTROPY_LOSS_BACK: ggml_op = 75;
-pub const ggml_op_GGML_OP_COUNT: ggml_op = 76;
 pub type ggml_op = ::std::os::raw::c_uint;
 pub const ggml_unary_op_GGML_UNARY_OP_ABS: ggml_unary_op = 0;
 pub const ggml_unary_op_GGML_UNARY_OP_SGN: ggml_unary_op = 1;
@@ -1614,15 +1631,11 @@ pub struct ggml_tensor {
     pub flags: i32,
     pub grad: *mut ggml_tensor,
     pub src: [*mut ggml_tensor; 10usize],
-    pub perf_runs: ::std::os::raw::c_int,
-    pub perf_cycles: i64,
-    pub perf_time_us: i64,
     pub view_src: *mut ggml_tensor,
     pub view_offs: usize,
     pub data: *mut ::std::os::raw::c_void,
     pub name: [::std::os::raw::c_char; 64usize],
     pub extra: *mut ::std::os::raw::c_void,
-    pub padding: [::std::os::raw::c_char; 8usize],
 }
 #[test]
 fn bindgen_test_layout_ggml_tensor() {
@@ -1630,7 +1643,7 @@ fn bindgen_test_layout_ggml_tensor() {
     let ptr = UNINIT.as_ptr();
     assert_eq!(
         ::std::mem::size_of::<ggml_tensor>(),
-        368usize,
+        336usize,
         concat!("Size of: ", stringify!(ggml_tensor))
     );
     assert_eq!(
@@ -1738,39 +1751,9 @@ fn bindgen_test_layout_ggml_tensor() {
             stringify!(src)
         )
     );
-    assert_eq!(
-        unsafe { ::std::ptr::addr_of!((*ptr).perf_runs) as usize - ptr as usize },
-        240usize,
-        concat!(
-            "Offset of field: ",
-            stringify!(ggml_tensor),
-            "::",
-            stringify!(perf_runs)
-        )
-    );
-    assert_eq!(
-        unsafe { ::std::ptr::addr_of!((*ptr).perf_cycles) as usize - ptr as usize },
-        248usize,
-        concat!(
-            "Offset of field: ",
-            stringify!(ggml_tensor),
-            "::",
-            stringify!(perf_cycles)
-        )
-    );
-    assert_eq!(
-        unsafe { ::std::ptr::addr_of!((*ptr).perf_time_us) as usize - ptr as usize },
-        256usize,
-        concat!(
-            "Offset of field: ",
-            stringify!(ggml_tensor),
-            "::",
-            stringify!(perf_time_us)
-        )
-    );
     assert_eq!(
         unsafe { ::std::ptr::addr_of!((*ptr).view_src) as usize - ptr as usize },
-        264usize,
+        240usize,
         concat!(
             "Offset of field: ",
             stringify!(ggml_tensor),
@@ -1780,7 +1763,7 @@ fn bindgen_test_layout_ggml_tensor() {
     );
     assert_eq!(
         unsafe { ::std::ptr::addr_of!((*ptr).view_offs) as usize - ptr as usize },
-        272usize,
+        248usize,
         concat!(
             "Offset of field: ",
             stringify!(ggml_tensor),
@@ -1790,7 +1773,7 @@ fn bindgen_test_layout_ggml_tensor() {
     );
     assert_eq!(
         unsafe { ::std::ptr::addr_of!((*ptr).data) as usize - ptr as usize },
-        280usize,
+        256usize,
         concat!(
             "Offset of field: ",
             stringify!(ggml_tensor),
@@ -1800,7 +1783,7 @@ fn bindgen_test_layout_ggml_tensor() {
     );
     assert_eq!(
         unsafe { ::std::ptr::addr_of!((*ptr).name) as usize - ptr as usize },
-        288usize,
+        264usize,
         concat!(
             "Offset of field: ",
             stringify!(ggml_tensor),
@@ -1810,7 +1793,7 @@ fn bindgen_test_layout_ggml_tensor() {
     );
     assert_eq!(
         unsafe { ::std::ptr::addr_of!((*ptr).extra) as usize - ptr as usize },
-        352usize,
+        328usize,
         concat!(
             "Offset of field: ",
             stringify!(ggml_tensor),
@@ -1818,18 +1801,8 @@ fn bindgen_test_layout_ggml_tensor() {
             stringify!(extra)
         )
     );
-    assert_eq!(
-        unsafe { ::std::ptr::addr_of!((*ptr).padding) as usize - ptr as usize },
-        360usize,
-        concat!(
-            "Offset of field: ",
-            stringify!(ggml_tensor),
-            "::",
-            stringify!(padding)
-        )
-    );
 }
-pub const GGML_TENSOR_SIZE: usize = 368;
+pub const GGML_TENSOR_SIZE: usize = 336;
 pub type ggml_abort_callback =
     ::std::option::Option<unsafe extern "C" fn(data: *mut ::std::os::raw::c_void) -> bool>;
 #[repr(C)]
@@ -1910,10 +1883,12 @@ pub const ggml_cgraph_eval_order_GGML_CGRAPH_EVAL_ORDER_LEFT_TO_RIGHT: ggml_cgra
 pub const ggml_cgraph_eval_order_GGML_CGRAPH_EVAL_ORDER_RIGHT_TO_LEFT: ggml_cgraph_eval_order = 1;
 pub const ggml_cgraph_eval_order_GGML_CGRAPH_EVAL_ORDER_COUNT: ggml_cgraph_eval_order = 2;
 pub type ggml_cgraph_eval_order = ::std::os::raw::c_uint;
+pub type ggml_bitset_t = u32;
 #[repr(C)]
 #[derive(Debug, Copy, Clone)]
 pub struct ggml_hash_set {
     pub size: usize,
+    pub used: *mut ggml_bitset_t,
     pub keys: *mut *mut ggml_tensor,
 }
 #[test]
@@ -1922,7 +1897,7 @@ fn bindgen_test_layout_ggml_hash_set() {
     let ptr = UNINIT.as_ptr();
     assert_eq!(
         ::std::mem::size_of::<ggml_hash_set>(),
-        16usize,
+        24usize,
         concat!("Size of: ", stringify!(ggml_hash_set))
     );
     assert_eq!(
@@ -1941,8 +1916,18 @@ fn bindgen_test_layout_ggml_hash_set() {
         )
     );
     assert_eq!(
-        unsafe { ::std::ptr::addr_of!((*ptr).keys) as usize - ptr as usize },
+        unsafe { ::std::ptr::addr_of!((*ptr).used) as usize - ptr as usize },
         8usize,
+        concat!(
+            "Offset of field: ",
+            stringify!(ggml_hash_set),
+            "::",
+            stringify!(used)
+        )
+    );
+    assert_eq!(
+        unsafe { ::std::ptr::addr_of!((*ptr).keys) as usize - ptr as usize },
+        16usize,
         concat!(
             "Offset of field: ",
             stringify!(ggml_hash_set),
@@ -1960,11 +1945,8 @@ pub struct ggml_cgraph {
     pub nodes: *mut *mut ggml_tensor,
     pub grads: *mut *mut ggml_tensor,
     pub leafs: *mut *mut ggml_tensor,
-    pub visited_hash_table: ggml_hash_set,
+    pub visited_hash_set: ggml_hash_set,
     pub order: ggml_cgraph_eval_order,
-    pub perf_runs: ::std::os::raw::c_int,
-    pub perf_cycles: i64,
-    pub perf_time_us: i64,
 }
 #[test]
 fn bindgen_test_layout_ggml_cgraph() {
@@ -1972,7 +1954,7 @@ fn bindgen_test_layout_ggml_cgraph() {
     let ptr = UNINIT.as_ptr();
     assert_eq!(
         ::std::mem::size_of::<ggml_cgraph>(),
-        80usize,
+        72usize,
         concat!("Size of: ", stringify!(ggml_cgraph))
     );
     assert_eq!(
@@ -2041,53 +2023,23 @@ fn bindgen_test_layout_ggml_cgraph() {
         )
     );
     assert_eq!(
-        unsafe { ::std::ptr::addr_of!((*ptr).visited_hash_table) as usize - ptr as usize },
+        unsafe { ::std::ptr::addr_of!((*ptr).visited_hash_set) as usize - ptr as usize },
         40usize,
         concat!(
             "Offset of field: ",
             stringify!(ggml_cgraph),
             "::",
-            stringify!(visited_hash_table)
+            stringify!(visited_hash_set)
         )
     );
     assert_eq!(
         unsafe { ::std::ptr::addr_of!((*ptr).order) as usize - ptr as usize },
-        56usize,
-        concat!(
-            "Offset of field: ",
-            stringify!(ggml_cgraph),
-            "::",
-            stringify!(order)
-        )
-    );
-    assert_eq!(
-        unsafe { ::std::ptr::addr_of!((*ptr).perf_runs) as usize - ptr as usize },
-        60usize,
-        concat!(
-            "Offset of field: ",
-            stringify!(ggml_cgraph),
-            "::",
-            stringify!(perf_runs)
-        )
-    );
-    assert_eq!(
-        unsafe { ::std::ptr::addr_of!((*ptr).perf_cycles) as usize - ptr as usize },
         64usize,
         concat!(
             "Offset of field: ",
             stringify!(ggml_cgraph),
             "::",
-            stringify!(perf_cycles)
+            stringify!(order)
-        )
-    );
-    assert_eq!(
-        unsafe { ::std::ptr::addr_of!((*ptr).perf_time_us) as usize - ptr as usize },
-        72usize,
-        concat!(
-            "Offset of field: ",
-            stringify!(ggml_cgraph),
-            "::",
-            stringify!(perf_time_us)
         )
     );
 }
@@ -2195,84 +2147,6 @@ fn bindgen_test_layout_ggml_init_params() {
         )
     );
 }
-pub const ggml_task_type_GGML_TASK_TYPE_INIT: ggml_task_type = 0;
-pub const ggml_task_type_GGML_TASK_TYPE_COMPUTE: ggml_task_type = 1;
-pub const ggml_task_type_GGML_TASK_TYPE_FINALIZE: ggml_task_type = 2;
-pub type ggml_task_type = ::std::os::raw::c_uint;
-#[repr(C)]
-#[derive(Debug, Copy, Clone)]
-pub struct ggml_compute_params {
-    pub type_: ggml_task_type,
-    pub ith: ::std::os::raw::c_int,
-    pub nth: ::std::os::raw::c_int,
-    pub wsize: usize,
-    pub wdata: *mut ::std::os::raw::c_void,
-}
-#[test]
-fn bindgen_test_layout_ggml_compute_params() {
-    const UNINIT: ::std::mem::MaybeUninit<ggml_compute_params> = ::std::mem::MaybeUninit::uninit();
-    let ptr = UNINIT.as_ptr();
-    assert_eq!(
-        ::std::mem::size_of::<ggml_compute_params>(),
-        32usize,
-        concat!("Size of: ", stringify!(ggml_compute_params))
-    );
-    assert_eq!(
-        ::std::mem::align_of::<ggml_compute_params>(),
-        8usize,
-        concat!("Alignment of ", stringify!(ggml_compute_params))
-    );
-    assert_eq!(
-        unsafe { ::std::ptr::addr_of!((*ptr).type_) as usize - ptr as usize },
-        0usize,
-        concat!(
-            "Offset of field: ",
-            stringify!(ggml_compute_params),
-            "::",
-            stringify!(type_)
-        )
-    );
-    assert_eq!(
-        unsafe { ::std::ptr::addr_of!((*ptr).ith) as usize - ptr as usize },
-        4usize,
-        concat!(
-            "Offset of field: ",
-            stringify!(ggml_compute_params),
-            "::",
-            stringify!(ith)
-        )
-    );
-    assert_eq!(
-        unsafe { ::std::ptr::addr_of!((*ptr).nth) as usize - ptr as usize },
-        8usize,
-        concat!(
-            "Offset of field: ",
-            stringify!(ggml_compute_params),
-            "::",
-            stringify!(nth)
-        )
-    );
-    assert_eq!(
-        unsafe { ::std::ptr::addr_of!((*ptr).wsize) as usize - ptr as usize },
-        16usize,
-        concat!(
-            "Offset of field: ",
-            stringify!(ggml_compute_params),
-            "::",
-            stringify!(wsize)
-        )
-    );
-    assert_eq!(
-        unsafe { ::std::ptr::addr_of!((*ptr).wdata) as usize - ptr as usize },
-        24usize,
-        concat!(
-            "Offset of field: ",
-            stringify!(ggml_compute_params),
-            "::",
-            stringify!(wdata)
-        )
-    );
-}
 pub const ggml_numa_strategy_GGML_NUMA_STRATEGY_DISABLED: ggml_numa_strategy = 0;
 pub const ggml_numa_strategy_GGML_NUMA_STRATEGY_DISTRIBUTE: ggml_numa_strategy = 1;
 pub const ggml_numa_strategy_GGML_NUMA_STRATEGY_ISOLATE: ggml_numa_strategy = 2;
@@ -2300,9 +2174,6 @@ extern "C" {
 extern "C" {
     pub fn ggml_cycles_per_ms() -> i64;
 }
-extern "C" {
-    pub fn ggml_print_backtrace();
-}
 extern "C" {
     pub fn ggml_fopen(
         fname: *const ::std::os::raw::c_char,
@@ -2334,7 +2205,7 @@ extern "C" {
     pub fn ggml_nbytes_pad(tensor: *const ggml_tensor) -> usize;
 }
 extern "C" {
-    pub fn ggml_blck_size(type_: ggml_type) -> ::std::os::raw::c_int;
+    pub fn ggml_blck_size(type_: ggml_type) -> i64;
 }
 extern "C" {
     pub fn ggml_type_size(type_: ggml_type) -> usize;
@@ -2372,9 +2243,6 @@ extern "C" {
 extern "C" {
     pub fn ggml_is_transposed(tensor: *const ggml_tensor) -> bool;
 }
-extern "C" {
-    pub fn ggml_is_contiguous(tensor: *const ggml_tensor) -> bool;
-}
 extern "C" {
     pub fn ggml_is_permuted(tensor: *const ggml_tensor) -> bool;
 }
@@ -2396,12 +2264,27 @@ extern "C" {
 extern "C" {
     pub fn ggml_n_dims(tensor: *const ggml_tensor) -> ::std::os::raw::c_int;
 }
+extern "C" {
+    pub fn ggml_is_contiguous(tensor: *const ggml_tensor) -> bool;
+}
+extern "C" {
+    pub fn ggml_is_contiguous_0(tensor: *const ggml_tensor) -> bool;
+}
+extern "C" {
+    pub fn ggml_is_contiguous_1(tensor: *const ggml_tensor) -> bool;
+}
+extern "C" {
+    pub fn ggml_is_contiguous_2(tensor: *const ggml_tensor) -> bool;
+}
 extern "C" {
     pub fn ggml_are_same_shape(t0: *const ggml_tensor, t1: *const ggml_tensor) -> bool;
 }
 extern "C" {
     pub fn ggml_are_same_stride(t0: *const ggml_tensor, t1: *const ggml_tensor) -> bool;
 }
+extern "C" {
+    pub fn ggml_can_repeat(t0: *const ggml_tensor, t1: *const ggml_tensor) -> bool;
+}
 extern "C" {
     pub fn ggml_tensor_overhead() -> usize;
 }
@@ -2757,6 +2640,7 @@ extern "C" {
         ctx: *mut ggml_context,
         a: *mut ggml_tensor,
         b: *mut ggml_tensor,
+        dim: ::std::os::raw::c_int,
     ) -> *mut ggml_tensor;
 }
 extern "C" {
@@ -2873,6 +2757,7 @@ extern "C" {
         ctx: *mut ggml_context,
         a: *mut ggml_tensor,
         n_groups: ::std::os::raw::c_int,
+        eps: f32,
     ) -> *mut ggml_tensor;
 }
 extern "C" {
@@ -2880,6 +2765,7 @@ extern "C" {
         ctx: *mut ggml_context,
         a: *mut ggml_tensor,
         n_groups: ::std::os::raw::c_int,
+        eps: f32,
     ) -> *mut ggml_tensor;
 }
 extern "C" {
@@ -3208,7 +3094,6 @@ extern "C" {
         b: *mut ggml_tensor,
         n_dims: ::std::os::raw::c_int,
         mode: ::std::os::raw::c_int,
-        n_ctx: ::std::os::raw::c_int,
     ) -> *mut ggml_tensor;
 }
 extern "C" {
@@ -3218,7 +3103,40 @@ extern "C" {
         b: *mut ggml_tensor,
         n_dims: ::std::os::raw::c_int,
         mode: ::std::os::raw::c_int,
-        n_ctx: ::std::os::raw::c_int,
+    ) -> *mut ggml_tensor;
+}
+extern "C" {
+    pub fn ggml_rope_ext(
+        ctx: *mut ggml_context,
+        a: *mut ggml_tensor,
+        b: *mut ggml_tensor,
+        c: *mut ggml_tensor,
+        n_dims: ::std::os::raw::c_int,
+        mode: ::std::os::raw::c_int,
+        n_ctx_orig: ::std::os::raw::c_int,
+        freq_base: f32,
+        freq_scale: f32,
+        ext_factor: f32,
+        attn_factor: f32,
+        beta_fast: f32,
+        beta_slow: f32,
+    ) -> *mut ggml_tensor;
+}
+extern "C" {
+    pub fn ggml_rope_ext_inplace(
+        ctx: *mut ggml_context,
+        a: *mut ggml_tensor,
+        b: *mut ggml_tensor,
+        c: *mut ggml_tensor,
+        n_dims: ::std::os::raw::c_int,
+        mode: ::std::os::raw::c_int,
+        n_ctx_orig: ::std::os::raw::c_int,
+        freq_base: f32,
+        freq_scale: f32,
+        ext_factor: f32,
+        attn_factor: f32,
+        beta_fast: f32,
+        beta_slow: f32,
     ) -> *mut ggml_tensor;
 }
 extern "C" {
@@ -3228,8 +3146,7 @@ extern "C" {
         b: *mut ggml_tensor,
         n_dims: ::std::os::raw::c_int,
         mode: ::std::os::raw::c_int,
-        n_ctx: ::std::os::raw::c_int,
+        n_ctx_orig: ::std::os::raw::c_int,
-        n_orig_ctx: ::std::os::raw::c_int,
         freq_base: f32,
         freq_scale: f32,
         ext_factor: f32,
@@ -3245,8 +3162,7 @@ extern "C" {
         b: *mut ggml_tensor,
         n_dims: ::std::os::raw::c_int,
         mode: ::std::os::raw::c_int,
-        n_ctx: ::std::os::raw::c_int,
+        n_ctx_orig: ::std::os::raw::c_int,
-        n_orig_ctx: ::std::os::raw::c_int,
         freq_base: f32,
         freq_scale: f32,
         ext_factor: f32,
@@ -3258,40 +3174,28 @@ extern "C" {
 extern "C" {
     pub fn ggml_rope_yarn_corr_dims(
         n_dims: ::std::os::raw::c_int,
-        n_orig_ctx: ::std::os::raw::c_int,
+        n_ctx_orig: ::std::os::raw::c_int,
         freq_base: f32,
         beta_fast: f32,
         beta_slow: f32,
         dims: *mut f32,
     );
 }
-extern "C" {
-    pub fn ggml_rope_xpos_inplace(
-        ctx: *mut ggml_context,
-        a: *mut ggml_tensor,
-        b: *mut ggml_tensor,
-        n_dims: ::std::os::raw::c_int,
-        base: f32,
-        down: bool,
-    ) -> *mut ggml_tensor;
-}
 extern "C" {
     pub fn ggml_rope_back(
         ctx: *mut ggml_context,
         a: *mut ggml_tensor,
         b: *mut ggml_tensor,
+        c: *mut ggml_tensor,
         n_dims: ::std::os::raw::c_int,
         mode: ::std::os::raw::c_int,
-        n_ctx: ::std::os::raw::c_int,
+        n_ctx_orig: ::std::os::raw::c_int,
-        n_orig_ctx: ::std::os::raw::c_int,
         freq_base: f32,
         freq_scale: f32,
         ext_factor: f32,
         attn_factor: f32,
         beta_fast: f32,
         beta_slow: f32,
-        xpos_base: f32,
-        xpos_down: bool,
     ) -> *mut ggml_tensor;
 }
 extern "C" {
@@ -3428,6 +3332,16 @@ extern "C" {
         scale_factor: ::std::os::raw::c_int,
     ) -> *mut ggml_tensor;
 }
+extern "C" {
+    pub fn ggml_upscale_ext(
+        ctx: *mut ggml_context,
+        a: *mut ggml_tensor,
+        ne0: ::std::os::raw::c_int,
+        ne1: ::std::os::raw::c_int,
+        ne2: ::std::os::raw::c_int,
+        ne3: ::std::os::raw::c_int,
+    ) -> *mut ggml_tensor;
+}
 extern "C" {
     pub fn ggml_pad(
         ctx: *mut ggml_context,
@@ -3471,15 +3385,6 @@ extern "C" {
         k: ::std::os::raw::c_int,
     ) -> *mut ggml_tensor;
 }
-extern "C" {
-    pub fn ggml_flash_attn(
-        ctx: *mut ggml_context,
-        q: *mut ggml_tensor,
-        k: *mut ggml_tensor,
-        v: *mut ggml_tensor,
-        masked: bool,
-    ) -> *mut ggml_tensor;
-}
 extern "C" {
     pub fn ggml_flash_attn_ext(
         ctx: *mut ggml_context,
@@ -3504,16 +3409,6 @@ extern "C" {
         masked: bool,
     ) -> *mut ggml_tensor;
 }
-extern "C" {
-    pub fn ggml_flash_ff(
-        ctx: *mut ggml_context,
-        a: *mut ggml_tensor,
-        b0: *mut ggml_tensor,
-        b1: *mut ggml_tensor,
-        c0: *mut ggml_tensor,
-        c1: *mut ggml_tensor,
-    ) -> *mut ggml_tensor;
-}
 extern "C" {
     pub fn ggml_ssm_conv(
         ctx: *mut ggml_context,
@@ -5115,12 +5010,18 @@ extern "C" {
 extern "C" {
     pub fn ggml_cpu_has_avx512_vnni() -> ::std::os::raw::c_int;
 }
+extern "C" {
+    pub fn ggml_cpu_has_avx512_bf16() -> ::std::os::raw::c_int;
+}
 extern "C" {
     pub fn ggml_cpu_has_fma() -> ::std::os::raw::c_int;
 }
 extern "C" {
     pub fn ggml_cpu_has_neon() -> ::std::os::raw::c_int;
 }
+extern "C" {
+    pub fn ggml_cpu_has_sve() -> ::std::os::raw::c_int;
+}
 extern "C" {
     pub fn ggml_cpu_has_arm_fma() -> ::std::os::raw::c_int;
 }
@@ -5142,9 +5043,6 @@ extern "C" {
 extern "C" {
     pub fn ggml_cpu_has_cuda() -> ::std::os::raw::c_int;
 }
-extern "C" {
-    pub fn ggml_cpu_has_clblast() -> ::std::os::raw::c_int;
-}
 extern "C" {
     pub fn ggml_cpu_has_vulkan() -> ::std::os::raw::c_int;
 }
@@ -5163,18 +5061,30 @@ extern "C" {
 extern "C" {
     pub fn ggml_cpu_has_sycl() -> ::std::os::raw::c_int;
 }
+extern "C" {
+    pub fn ggml_cpu_has_rpc() -> ::std::os::raw::c_int;
+}
 extern "C" {
     pub fn ggml_cpu_has_vsx() -> ::std::os::raw::c_int;
 }
 extern "C" {
     pub fn ggml_cpu_has_matmul_int8() -> ::std::os::raw::c_int;
 }
+extern "C" {
+    pub fn ggml_cpu_has_cann() -> ::std::os::raw::c_int;
+}
+extern "C" {
+    pub fn ggml_cpu_has_llamafile() -> ::std::os::raw::c_int;
+}
 pub type ggml_to_float_t = ::std::option::Option<
     unsafe extern "C" fn(x: *const ::std::os::raw::c_void, y: *mut f32, k: i64),
 >;
 pub type ggml_from_float_t = ::std::option::Option<
     unsafe extern "C" fn(x: *const f32, y: *mut ::std::os::raw::c_void, k: i64),
 >;
+pub type ggml_from_float_to_mat_t = ::std::option::Option<
+    unsafe extern "C" fn(x: *const f32, y: *mut ::std::os::raw::c_void, nr: i64, k: i64, bs: i64),
+>;
 pub type ggml_vec_dot_t = ::std::option::Option<
     unsafe extern "C" fn(
         n: ::std::os::raw::c_int,
@@ -5187,19 +5097,46 @@ pub type ggml_vec_dot_t = ::std::option::Option<
         nrc: ::std::os::raw::c_int,
     ),
 >;
+pub type ggml_gemv_t = ::std::option::Option<
+    unsafe extern "C" fn(
+        n: ::std::os::raw::c_int,
+        s: *mut f32,
+        bs: usize,
+        x: *const ::std::os::raw::c_void,
+        y: *const ::std::os::raw::c_void,
+        nr: ::std::os::raw::c_int,
+        nc: ::std::os::raw::c_int,
+    ),
+>;
+pub type ggml_gemm_t = ::std::option::Option<
+    unsafe extern "C" fn(
+        n: ::std::os::raw::c_int,
+        s: *mut f32,
+        bs: usize,
+        x: *const ::std::os::raw::c_void,
+        y: *const ::std::os::raw::c_void,
+        nr: ::std::os::raw::c_int,
+        nc: ::std::os::raw::c_int,
+    ),
+>;
 #[repr(C)]
 #[derive(Debug, Copy, Clone)]
 pub struct ggml_type_traits_t {
     pub type_name: *const ::std::os::raw::c_char,
-    pub blck_size: ::std::os::raw::c_int,
+    pub blck_size: i64,
+    pub blck_size_interleave: i64,
     pub type_size: usize,
     pub is_quantized: bool,
     pub to_float: ggml_to_float_t,
     pub from_float: ggml_from_float_t,
-    pub from_float_reference: ggml_from_float_t,
+    pub from_float_ref: ggml_from_float_t,
+    pub from_float_to_mat: ggml_from_float_to_mat_t,
     pub vec_dot: ggml_vec_dot_t,
     pub vec_dot_type: ggml_type,
     pub nrows: i64,
+    pub ncols: i64,
+    pub gemv: ggml_gemv_t,
+    pub gemm: ggml_gemm_t,
 }
 #[test]
 fn bindgen_test_layout_ggml_type_traits_t() {
@@ -5207,7 +5144,7 @@ fn bindgen_test_layout_ggml_type_traits_t() {
     let ptr = UNINIT.as_ptr();
     assert_eq!(
         ::std::mem::size_of::<ggml_type_traits_t>(),
-        80usize,
+        120usize,
         concat!("Size of: ", stringify!(ggml_type_traits_t))
     );
     assert_eq!(
@@ -5236,8 +5173,18 @@ fn bindgen_test_layout_ggml_type_traits_t() {
         )
     );
     assert_eq!(
-        unsafe { ::std::ptr::addr_of!((*ptr).type_size) as usize - ptr as usize },
+        unsafe { ::std::ptr::addr_of!((*ptr).blck_size_interleave) as usize - ptr as usize },
         16usize,
+        concat!(
+            "Offset of field: ",
+            stringify!(ggml_type_traits_t),
+            "::",
+            stringify!(blck_size_interleave)
+        )
+    );
+    assert_eq!(
+        unsafe { ::std::ptr::addr_of!((*ptr).type_size) as usize - ptr as usize },
+        24usize,
         concat!(
             "Offset of field: ",
             stringify!(ggml_type_traits_t),
@@ -5247,7 +5194,7 @@ fn bindgen_test_layout_ggml_type_traits_t() {
     );
     assert_eq!(
         unsafe { ::std::ptr::addr_of!((*ptr).is_quantized) as usize - ptr as usize },
-        24usize,
+        32usize,
         concat!(
             "Offset of field: ",
             stringify!(ggml_type_traits_t),
@@ -5257,7 +5204,7 @@ fn bindgen_test_layout_ggml_type_traits_t() {
     );
     assert_eq!(
         unsafe { ::std::ptr::addr_of!((*ptr).to_float) as usize - ptr as usize },
-        32usize,
+        40usize,
         concat!(
             "Offset of field: ",
             stringify!(ggml_type_traits_t),
@@ -5267,7 +5214,7 @@ fn bindgen_test_layout_ggml_type_traits_t() {
     );
     assert_eq!(
         unsafe { ::std::ptr::addr_of!((*ptr).from_float) as usize - ptr as usize },
-        40usize,
+        48usize,
         concat!(
             "Offset of field: ",
             stringify!(ggml_type_traits_t),
@@ -5276,18 +5223,28 @@ fn bindgen_test_layout_ggml_type_traits_t() {
         )
     );
     assert_eq!(
-        unsafe { ::std::ptr::addr_of!((*ptr).from_float_reference) as usize - ptr as usize },
+        unsafe { ::std::ptr::addr_of!((*ptr).from_float_ref) as usize - ptr as usize },
-        48usize,
+        56usize,
         concat!(
             "Offset of field: ",
             stringify!(ggml_type_traits_t),
             "::",
-            stringify!(from_float_reference)
+            stringify!(from_float_ref)
+        )
+    );
+    assert_eq!(
+        unsafe { ::std::ptr::addr_of!((*ptr).from_float_to_mat) as usize - ptr as usize },
+        64usize,
+        concat!(
+            "Offset of field: ",
+            stringify!(ggml_type_traits_t),
+            "::",
+            stringify!(from_float_to_mat)
         )
     );
     assert_eq!(
         unsafe { ::std::ptr::addr_of!((*ptr).vec_dot) as usize - ptr as usize },
-        56usize,
+        72usize,
         concat!(
             "Offset of field: ",
             stringify!(ggml_type_traits_t),
@@ -5297,7 +5254,7 @@ fn bindgen_test_layout_ggml_type_traits_t() {
     );
     assert_eq!(
         unsafe { ::std::ptr::addr_of!((*ptr).vec_dot_type) as usize - ptr as usize },
-        64usize,
+        80usize,
         concat!(
             "Offset of field: ",
             stringify!(ggml_type_traits_t),
@@ -5307,7 +5264,7 @@ fn bindgen_test_layout_ggml_type_traits_t() {
     );
     assert_eq!(
         unsafe { ::std::ptr::addr_of!((*ptr).nrows) as usize - ptr as usize },
-        72usize,
+        88usize,
         concat!(
             "Offset of field: ",
             stringify!(ggml_type_traits_t),
@@ -5315,6 +5272,36 @@ fn bindgen_test_layout_ggml_type_traits_t() {
             stringify!(nrows)
         )
     );
+    assert_eq!(
+        unsafe { ::std::ptr::addr_of!((*ptr).ncols) as usize - ptr as usize },
+        96usize,
+        concat!(
+            "Offset of field: ",
+            stringify!(ggml_type_traits_t),
+            "::",
+            stringify!(ncols)
+        )
+    );
+    assert_eq!(
+        unsafe { ::std::ptr::addr_of!((*ptr).gemv) as usize - ptr as usize },
+        104usize,
+        concat!(
+            "Offset of field: ",
+            stringify!(ggml_type_traits_t),
+            "::",
+            stringify!(gemv)
+        )
+    );
+    assert_eq!(
+        unsafe { ::std::ptr::addr_of!((*ptr).gemm) as usize - ptr as usize },
+        112usize,
+        concat!(
+            "Offset of field: ",
+            stringify!(ggml_type_traits_t),
+            "::",
+            stringify!(gemm)
+        )
+    );
 }
 extern "C" {
     pub fn ggml_internal_get_type_traits(type_: ggml_type) -> ggml_type_traits_t;
@@ -5897,23 +5884,6 @@ extern "C" {
         n_threads: ::std::os::raw::c_int,
     ) -> ::std::os::raw::c_int;
 }
-extern "C" {
-    pub fn whisper_pcm_to_mel_phase_vocoder(
-        ctx: *mut whisper_context,
-        samples: *const f32,
-        n_samples: ::std::os::raw::c_int,
-        n_threads: ::std::os::raw::c_int,
-    ) -> ::std::os::raw::c_int;
-}
-extern "C" {
-    pub fn whisper_pcm_to_mel_phase_vocoder_with_state(
-        ctx: *mut whisper_context,
-        state: *mut whisper_state,
-        samples: *const f32,
-        n_samples: ::std::os::raw::c_int,
-        n_threads: ::std::os::raw::c_int,
-    ) -> ::std::os::raw::c_int;
-}
 extern "C" {
     pub fn whisper_set_mel(
         ctx: *mut whisper_context,
@@ -6177,7 +6147,6 @@ pub struct whisper_full_params {
     pub max_len: ::std::os::raw::c_int,
     pub split_on_word: bool,
     pub max_tokens: ::std::os::raw::c_int,
-    pub speed_up: bool,
     pub debug_mode: bool,
     pub audio_ctx: ::std::os::raw::c_int,
     pub tdrz_enable: bool,
@@ -6496,19 +6465,9 @@ fn bindgen_test_layout_whisper_full_params() {
             stringify!(max_tokens)
         )
     );
-    assert_eq!(
-        unsafe { ::std::ptr::addr_of!((*ptr).speed_up) as usize - ptr as usize },
-        52usize,
-        concat!(
-            "Offset of field: ",
-            stringify!(whisper_full_params),
-            "::",
-            stringify!(speed_up)
-        )
-    );
     assert_eq!(
         unsafe { ::std::ptr::addr_of!((*ptr).debug_mode) as usize - ptr as usize },
-        53usize,
+        52usize,
         concat!(
             "Offset of field: ",
             stringify!(whisper_full_params),

sys/whisper.cpp

@@ -1 +1 @@
-Subproject commit c7b6988678779901d02ceba1a8212d2c9908956e
+Subproject commit 9e3c5345cd46ea718209db53464e426c3fe7a25e