Update whisper.cpp version to 1.6.2

Cargo.toml

@@ -4,7 +4,7 @@ exclude = ["examples/full_usage"]
 
 [package]
 name = "whisper-rs"
-version = "0.11.0"
+version = "0.12.0"
 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 = "0.8" }
+whisper-rs-sys = { path = "sys", version = "0.10.0" }
 log = { version = "0.4", optional = true }
 tracing = { version = "0.1", optional = true }
 

src/lib.rs

@@ -44,5 +44,5 @@ pub type WhisperNewSegmentCallback = whisper_rs_sys::whisper_new_segment_callbac
 pub type WhisperStartEncoderCallback = whisper_rs_sys::whisper_encoder_begin_callback;
 pub type WhisperProgressCallback = whisper_rs_sys::whisper_progress_callback;
 pub type WhisperLogitsFilterCallback = whisper_rs_sys::whisper_logits_filter_callback;
-pub type WhisperAbortCallback = whisper_rs_sys::whisper_abort_callback;
+pub type WhisperAbortCallback = whisper_rs_sys::ggml_abort_callback;
 pub type WhisperLogCallback = whisper_rs_sys::ggml_log_callback;

src/standalone.rs

@@ -105,7 +105,7 @@ pub struct SystemInfo {
     pub f16c: bool,
     pub blas: bool,
     pub clblast: bool,
-    pub cublas: bool,
+    pub cuda: bool,
 }
 
 impl Default for SystemInfo {
@@ -118,7 +118,7 @@ impl Default for SystemInfo {
                 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,
-                cublas: whisper_rs_sys::ggml_cpu_has_cublas() != 0,
+                cuda: whisper_rs_sys::ggml_cpu_has_cuda() != 0,
             }
         }
     }

src/whisper_ctx.rs

@@ -475,13 +475,37 @@ pub struct WhisperContextParameters {
     /// **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
+    pub flash_attn : bool,
+    /// GPU device id, default 0
+    pub gpu_device: c_int,
+    /// [EXPERIMENTAL] Enable Token-level timestamps with DTW, default 0
+    pub dtw_token_timestamps: bool,
+    /// Preset id for DTW, default whisper_alignment_heads_preset_WHISPER_AHEADS_NONE
+    pub dtw_aheads_preset : whisper_rs_sys::whisper_alignment_heads_preset,
+    /// Number of top text layers used from model. Only with whisper_alignment_heads_preset_WHISPER_AHEADS_N_TOP_MOST preset.
+    pub dtw_n_top : c_int,
+    /// Custom aheads, only with whisper_alignment_heads_preset_WHISPER_AHEADS_CUSTOM preset
+    /// See details https://github.com/ggerganov/whisper.cpp/pull/1485#discussion_r1519681143
+    pub dtw_aheads: whisper_rs_sys::whisper_aheads,
+    /// Memory size for DTW
+    ///
+    /// **Warning**: Might be removed in next version of whisper.cpp
+    pub dtw_mem_size : usize
 }
 
 #[allow(clippy::derivable_impls)] // this impl cannot be derived
 impl Default for WhisperContextParameters {
     fn default() -> Self {
         Self {
             use_gpu: cfg!(feature = "_gpu"),
+            flash_attn : false,
+            gpu_device: 0,
+            dtw_token_timestamps: false,
+            dtw_aheads_preset : whisper_rs_sys::whisper_alignment_heads_preset_WHISPER_AHEADS_NONE,
+            dtw_n_top: -1,
+            dtw_aheads :  whisper_rs_sys::whisper_aheads { n_heads: 0, heads: std::ptr::null() },
+            dtw_mem_size : 1024 * 1024 * 128
         }
     }
 }
@@ -493,9 +517,44 @@ impl WhisperContextParameters {
         self.use_gpu = use_gpu;
         self
     }
+    pub fn flash_attn(&mut self, flash_attn: bool) -> &mut Self {
+        self.flash_attn = flash_attn;
+        self
+    }
+    pub fn gpu_device(&mut self, gpu_device: c_int) -> &mut Self {
+        self.gpu_device = gpu_device;
+        self
+    }
+    pub fn dtw_token_timestamps(&mut self, dtw_token_timestamps: bool) -> &mut Self {
+        self.dtw_token_timestamps = dtw_token_timestamps;
+        self
+    }
+    pub fn dtw_aheads_preset(&mut self, dtw_aheads_preset: whisper_rs_sys::whisper_alignment_heads_preset) -> &mut Self {
+        self.dtw_aheads_preset = dtw_aheads_preset;
+        self
+    }
+    pub fn dtw_n_top(&mut self, dtw_n_top: c_int) -> &mut Self {
+        self.dtw_n_top = dtw_n_top;
+        self
+    }
+    pub fn dtw_aheads(&mut self, dtw_aheads: whisper_rs_sys::whisper_aheads) -> &mut Self {
+        self.dtw_aheads = dtw_aheads;
+        self
+    }
+    pub fn dtw_mem_size(&mut self, dtw_mem_size: usize) -> &mut Self {
+        self.dtw_mem_size = dtw_mem_size;
+        self
+    }
     fn to_c_struct(&self) -> whisper_rs_sys::whisper_context_params {
         whisper_rs_sys::whisper_context_params {
             use_gpu: self.use_gpu,
+            flash_attn: self.flash_attn,
+            gpu_device: self.gpu_device,
+            dtw_token_timestamps: self.dtw_token_timestamps,
+            dtw_aheads_preset: self.dtw_aheads_preset,
+            dtw_n_top: self.dtw_n_top,
+            dtw_aheads: self.dtw_aheads,
+            dtw_mem_size: self.dtw_mem_size,
         }
     }
 }

sys/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "whisper-rs-sys"
-version = "0.8.1"
+version = "0.10.0"
 edition = "2021"
 description = "Rust bindings for whisper.cpp (FFI bindings)"
 license = "Unlicense"

sys/build.rs

@@ -121,7 +121,7 @@ fn main() {
     }
 
     if cfg!(feature = "cuda") {
-        config.define("WHISPER_CUBLAS", "ON");
+        config.define("WHISPER_CUDA", "ON");
     }
 
     if cfg!(feature = "openblas") {