multiscreen-rs 0.3.2

A Rust implementation of the Multiscreen neural language model — training and inference powered by Burn.
Documentation 
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

// ---- High-level API (recommended for most users) ----
pub use crate::{
    ChatModel, GenerationConfig, ParameterBudget, Trainer, TrainingReport, auto_device, cuda,
};

#[cfg(not(feature = "cuda"))]
pub use crate::cpu;

// ---- Core types ----
pub use crate::{
    DefaultAutodiffBackend, DefaultBackend, DefaultMultiscreenModel, Device, Error, Result,
    device_label,
};

#[cfg(not(feature = "cuda"))]
pub use crate::default_device;

// ---- Model configuration ----
pub use crate::{
    EvaluationResult, ModelInferenceConfig, ModelTrainingConfig, ModelTrainingReport,
    MultiscreenModel, MultiscreenModelConfig, MultiscreenModelOutput, MultiscreenParameterBudget,
    cross_entropy_loss_with_mask,
};

// ---- Engine (lightweight transition engine) ----
pub use crate::{InferenceOutput, MultiscreenEngine, TrainInput, TrainReport};

// ---- Layout utilities ----
pub use crate::{
    InferenceConfig, Int, MultiscreenConfig, Screen, ScreenConfig, ScreenLayout,
    ScreeningGridConfig, Tensor, TensorData, TileConfig, TokenSpan, TrimConfig, causal_softmask,
    causal_trim_relevance, trim_and_square,
};

#[cfg(feature = "cuda")]
pub use crate::{CudaAutodiffBackend, CudaDevice, CudaMultiscreenModel};

pub use crate::AutodiffBackend;
pub use crate::Backend;