ruda-nn 0.21.62

Ruda neural network layers, activation modules and losses.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97

/// Attention module
pub mod attention;

/// Cache module
pub mod cache;

/// Convolution module
pub mod conv;

/// Pooling module
pub mod pool;

/// Transformer module
pub mod transformer;

/// Interpolate module
pub mod interpolate;

mod dropout;
mod embedding;
mod linear;
mod lora;
mod lora_record;
mod awq;
mod nf4;
mod quantized_linear;
mod nf4_experts;
mod nf4_expert_window;
mod packed_experts;
mod expert_lora;
mod floating_expert_lora;
mod mixed_expert_lora;
mod expert_lora_record;
mod expert_adapter_access;
mod moe;
/// Explicit tensor-parallel linear projections for replicated-loss model partitions.
#[cfg(feature = "tensor-parallel")]
pub mod tensor_parallel;
/// Native cross-rank routed experts with explicit global ownership and transport.
#[cfg(feature = "tensor-parallel")]
pub mod expert_parallel;
#[cfg(feature = "tensor-parallel")]
mod expert_parallel_adapters;
#[cfg(feature = "tensor-parallel")]
mod packed_expert_parallel;
#[cfg(feature="tensor-parallel")]
mod mixed_expert_parallel;
/// Element-sharded parameters with differentiable gather/reduce-scatter.
#[cfg(feature = "tensor-parallel")]
pub mod fully_sharded;
/// Joint data-sharded storage and tensor-parallel native projections.
#[cfg(feature = "tensor-parallel")]
pub mod hybrid_sharded;
/// Manifold-constrained hyper-connections and residual mixing.
pub mod mhc;
#[cfg(feature = "sparse")]
mod sparse_linear;
mod noise;
mod pos_encoding;
mod rnn;
mod rope_encoding;
mod unfold;

pub mod norm;
pub use norm::{batch::*, group::*, instance::*, layer::*, local_response::*, rms::*};

pub use dropout::*;
pub use embedding::*;
pub use linear::*;
pub use lora::*;
pub use lora_record::*;
pub use awq::*;
pub use nf4::*;
pub use quantized_linear::*;
pub use nf4_experts::*;
pub use nf4_expert_window::*;
pub use packed_experts::*;
pub use expert_lora::*;
pub use floating_expert_lora::*;
pub use mixed_expert_lora::*;
pub use expert_lora_record::*;
pub use expert_adapter_access::*;
#[cfg(feature = "tensor-parallel")]
pub use expert_parallel_adapters::*;
#[cfg(feature = "tensor-parallel")]
pub use packed_expert_parallel::*;
#[cfg(feature="tensor-parallel")]
pub use mixed_expert_parallel::*;
pub use moe::*;
pub use mhc::*;
#[cfg(feature = "sparse")]
pub use sparse_linear::*;
pub use noise::*;
pub use pos_encoding::*;
pub use rnn::*;
pub use rope_encoding::*;
pub use unfold::*;