Merge pull request #504 from brittlewis12/ubatch

Expose `n_ubatch` context param

llama-cpp-2/src/context.rs

@@ -52,12 +52,18 @@ impl<'model> LlamaContext<'model> {
         }
     }
 
-    /// Gets the max number of tokens in a batch.
+    /// Gets the max number of logical tokens that can be submitted to decode. Must be greater than or equal to n_ubatch.
     #[must_use]
     pub fn n_batch(&self) -> u32 {
         unsafe { llama_cpp_sys_2::llama_n_batch(self.context.as_ptr()) }
     }
 
+    /// Gets the max number of physical tokens (hardware level) to decode in batch. Must be less than or equal to n_batch.
+    #[must_use]
+    pub fn n_ubatch(&self) -> u32 {
+        unsafe { llama_cpp_sys_2::llama_n_ubatch(self.context.as_ptr()) }
+    }
+
     /// Gets the size of the context.
     #[must_use]
     pub fn n_ctx(&self) -> u32 {

llama-cpp-2/src/context/params.rs

@@ -166,6 +166,37 @@ impl LlamaContextParams {
         self.context_params.n_batch
     }
 
+    /// Set the `n_ubatch`
+    ///
+    /// # Examples
+    ///
+    /// ```rust
+    /// # use std::num::NonZeroU32;
+    /// use llama_cpp_2::context::params::LlamaContextParams;
+    /// let params = LlamaContextParams::default()
+    ///     .with_n_ubatch(512);
+    /// assert_eq!(params.n_ubatch(), 512);
+    /// ```
+    #[must_use]
+    pub fn with_n_ubatch(mut self, n_ubatch: u32) -> Self {
+        self.context_params.n_ubatch = n_ubatch;
+        self
+    }
+
+    /// Get the `n_ubatch`
+    ///
+    /// # Examples
+    ///
+    /// ```rust
+    /// use llama_cpp_2::context::params::LlamaContextParams;
+    /// let params = LlamaContextParams::default();
+    /// assert_eq!(params.n_ubatch(), 512);
+    /// ```
+    #[must_use]
+    pub fn n_ubatch(&self) -> u32 {
+        self.context_params.n_ubatch
+    }
+
     /// Set the type of rope scaling.
     ///
     /// # Examples