Crate ds_r1_rs

Crate ds_r1_rs 

Source
Expand description

§DeepSeek R1 Rust Implementation

A prototype implementation of DeepSeek R1-inspired reasoning model in Rust. This library provides core components for transformer architecture, multi-head latent attention (MLA), mixture-of-experts (MoE), and reasoning capabilities.

Re-exports§

pub use model::config::ModelConfig;
pub use model::transformer::DeepSeekR1Model;
pub use inference::engine::InferenceEngine;
pub use inference::reasoning::ReasoningOutput;
pub use training::data::TrainingExample;
pub use training::trainer::BasicTrainer;
pub use utils::error::ModelError;
pub use utils::error::Result;
pub use utils::math::MathUtils;

Modules§

inference
Inference Module
model
Model Module
training
Training Module
utils
Utilities Module

Constants§

VERSION
Library version

Functions§

default_config
Default model configuration for quick prototyping