god_graph/transformer/
mod.rs

1//! Transformer module for LLM support
2//!
3//! This module provides:
4//! - Autograd engine for automatic differentiation
5//! - Transformer layer implementations (Attention, FFN, Norm)
6//! - Model loading from Safetensors format
7//! - Graph-structured Transformer inference
8//! - KV Cache and batch inference optimizations
9//! - Sparse attention optimizations
10//! - Quantization support
11//! - Performance optimizations (SIMD, memory pool, optimized kernels)
12
13#![cfg(feature = "tensor")]
14
15pub mod autograd;
16pub mod layers;
17pub mod loader;
18pub mod model;
19pub mod generation;
20pub mod graph_transformer;
21pub mod kv_cache;
22#[cfg(feature = "tensor")]
23pub mod sparse_attention;
24pub mod batch;
25#[cfg(feature = "tensor")]
26pub mod quantization;
27pub mod perf;
28pub mod optimization;
29
30// Re-export commonly used types
31pub use autograd::{ComputeGraph, DifferentiableTensor, Op, Optimizer};
32pub use layers::{MultiHeadAttention, RMSNorm, LayerNorm, RoPE, FeedForward};
33pub use loader::{SafetensorsLoader, ModelConfig};
34pub use model::{LlamaModel, LlamaConfig};
35pub use generation::{GenerationConfig, TextGenerator};
36pub use graph_transformer::{GraphExecutor, GraphTransformer, GraphNode, GraphEdge};
37pub use kv_cache::KVCache;
38#[cfg(feature = "tensor")]
39pub use sparse_attention::SparseAttention;
40pub use batch::BatchInference;
41pub use quantization::{QuantizedTensor, QuantizationConfig};
42pub use perf::{TransformerMemoryPool, softmax_inplace_simd, matmul_with_buffer};
god_graph/transformer/mod.rs

god_graph/transformer/
mod.rs
