pub struct RadamAdaptiveStep {
pub name: &'static str,
pub id: String,
pub block_size: i32,
pub kernel_source: String,
pub entry_point_source: String,
pub source: String,
}Expand description
RAdam step — adaptive variance phase (rho_t > 5).
exp_avg = beta1 * exp_avg + (1 - beta1) * g
exp_avg_sq = beta2 * exp_avg_sq + (1 - beta2) * g²
denom = sqrt(exp_avg_sq / bias_correction2) + eps
p -= step_size * exp_avg / denomstep_size = lr * rect / bias_correction1 is precomputed on host, where
rect = sqrt((rho_t-4)(rho_t-2)*rho_inf / ((rho_inf-4)(rho_inf-2)*rho_t)).
Grid: [ceil(n_elements / BLOCK_SIZE), 1, 1].
Fields§
§name: &'static strThe kernel function’s name (e.g. "flash_attention2_forward").
id: StringUnique kernel identifier: fn_name + dtype(s) + const values joined by “__”.
block_size: i32Compile-time kernel constant, from the annotated fn’s const generics.
kernel_source: StringThe original kernel function source.
entry_point_source: StringThe Rust source of the generated C-ABI entry-point wrapper function.
source: StringCombined source (kernel_source + "\n\n" + entry_point_source); used by the Kernel trait.
Implementations§
Trait Implementations§
Source§impl Kernel for RadamAdaptiveStep
impl Kernel for RadamAdaptiveStep
Source§type Args<'__a> = (*mut f32, *mut f32, *mut f32, *mut f32, i32, f32, f32, f32, f32, f32, f32)
type Args<'__a> = (*mut f32, *mut f32, *mut f32, *mut f32, i32, f32, f32, f32, f32, f32, f32)
This kernel’s launch-argument tuple type.
Source§fn kernel_source(&self) -> &str
fn kernel_source(&self) -> &str
This kernel’s body source, without the entry-point wrapper.
Source§fn entry_point_source(&self) -> &str
fn entry_point_source(&self) -> &str
Returns the Rust source of the generated C-ABI entry-point wrapper function.
Source§fn entry_point_name(&self) -> String
fn entry_point_name(&self) -> String
Returns the PTX symbol name for this kernel:
"{name}_entry_point".Auto Trait Implementations§
impl Freeze for RadamAdaptiveStep
impl RefUnwindSafe for RadamAdaptiveStep
impl Send for RadamAdaptiveStep
impl Sync for RadamAdaptiveStep
impl Unpin for RadamAdaptiveStep
impl UnsafeUnpin for RadamAdaptiveStep
impl UnwindSafe for RadamAdaptiveStep
Blanket Implementations§
Source§impl<T> BorrowMut<T> for Twhere
T: ?Sized,
impl<T> BorrowMut<T> for Twhere
T: ?Sized,
Source§fn borrow_mut(&mut self) -> &mut T
fn borrow_mut(&mut self) -> &mut T
Mutably borrows from an owned value. Read more