boostr 0.2.0-beta.2

ML framework built on numr - attention, quantization, model architectures
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20

pub mod flash;
pub mod flash_standard;
pub mod fused_qkv;
pub mod mla;
pub mod paged;
#[path = "rope.rs"]
pub mod rope;

pub use flash::multi_head_attention_impl;
pub use flash_standard::{
    StandardAttnConfig, build_attention_mask, standard_attention_bwd, standard_attention_fwd,
    sum_gqa_grads,
};
pub use fused_qkv::{
    fused_output_projection_residual_bwd_impl, fused_output_projection_residual_impl,
    fused_qkv_projection_bwd_impl, fused_qkv_projection_impl,
};
pub use mla::scaled_dot_product_attention_impl;
pub use paged::{PagedAttnConfig, PagedKv, paged_attention_bwd_impl};
pub use rope::{apply_rope_impl, apply_rope_interleaved_impl, apply_rope_yarn_impl};