35 lines
1.0 KiB
Rust
35 lines
1.0 KiB
Rust
//! `LLaMA` Architecture - Main Module
|
|
//!
|
|
//! Re-exports from split modules for backward compatibility
|
|
|
|
// Re-export from attention module
|
|
pub use crate::architectures::llama_attention::{
|
|
LLaMAAttention, LLaMAConfig, RopeScaling, RotaryPositionEmbedding, SwiGLU,
|
|
};
|
|
|
|
// Re-export from model module (commented out until llama_model is implemented)
|
|
// pub use crate::architectures::llama_model::{
|
|
// LLaMAModel, LLaMABlock
|
|
// };
|
|
|
|
// Re-export tests for backward compatibility
|
|
#[cfg(all(test, feature = "disabled_tests"))]
|
|
mod tests {
|
|
use super::*;
|
|
use crate::architectures::llama_attention::LLaMAModel;
|
|
use rtx_tensor::Device;
|
|
|
|
#[test]
|
|
fn test_llama_integration() {
|
|
let config = LLaMAConfig::llama_7b();
|
|
let device = Device::cuda(0).unwrap_or(Device::default());
|
|
|
|
// Test complete integration
|
|
let model = LLaMAModel::new(config, &device);
|
|
assert!(model.is_ok());
|
|
|
|
let model = model.unwrap();
|
|
assert_eq!(model.architecture_type(), "LLaMA");
|
|
}
|
|
}
|