oarust 0.8.6

GPU-first semantic computing engine written in Rust
//! Reverse-mode differentiation lifecycle and private operation attachments.
//!
//! [`GradientTape`] is the only public autograd owner. Concrete nodes and
//! generated attachment policy remain private implementation details paired
//! with their owning operation families.

mod flow;
#[path = "autograd/loss.gen.rs"]
mod loss;
mod matrix;
mod node;
mod tape;

pub use tape::GradientTape;
pub(in crate::ml) use tape::{record_parameter_leaf, recording_active};

pub(in crate::ml) use flow::{
	record_flow_euler_step, record_flow_linear_match, record_flow_masked_mse,
};

pub(in crate::ml) use loss::{
	record_bce, record_cross_entropy, record_l1, record_masked_cross_entropy, record_mse,
	record_ppo_clipped_policy, record_smooth_l1,
};
pub(in crate::ml) use matrix::{
	record_adaptive_avg_pool_2d, record_avg_pool_2d, record_batch_norm_2d, record_bmm, record_bmm_nt,
	record_bmm_tn, record_channel_norm, record_conv_1d, record_conv_2d, record_conv_transpose_1d,
	record_conv_transpose_2d, record_elu, record_embedding, record_flash_attention, record_gelu,
	record_grouped_gemm_m, record_grouped_linear_m, record_gru_cell, record_gru_scan,
	record_layer_norm, record_leaky_relu, record_linear, record_mamba3_mimo,
	record_mamba3_preprocess, record_mamba3_siso, record_max_pool_2d, record_merge_heads,
	record_mish, record_moe_combine, record_moe_gather, record_moe_route_weights, record_relu,
	record_rms_norm, record_rms_norm_gated, record_rnn_cell, record_rnn_scan, record_rope,
	record_scaled_dot_product_attention, record_sigmoid, record_silu, record_silu_mul,
	record_softmax_scaled_masked, record_softplus, record_split_heads, record_swiglu, record_tanh,
	record_upsample_2d,
};