39 lines
1.6 KiB
Rust
39 lines
1.6 KiB
Rust
// SPDX-FileCopyrightText: Copyright (c) 2024-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
|
|
// SPDX-License-Identifier: Apache-2.0
|
|
|
|
/// This module provides a high-performance interface that bridges Python
|
|
/// applications with the Rust-powered Dynamo LLM runtime.
|
|
///
|
|
/// It is organized into several specialized sub-modules, each responsible for a particular aspect of the system:
|
|
///
|
|
/// - `backend`:
|
|
/// Wraps low-level interfaces for LLM inference, manages resource allocation,
|
|
/// and integrates with specialized hardware for optimized execution.
|
|
/// - `disagg_route`:
|
|
/// Implements distributed routing of inference requests with dynamic
|
|
/// load balancing and efficient resource allocation across clusters.
|
|
/// - `kv`:
|
|
/// Implements a high-performance key-value caching system that stores
|
|
/// intermediate computations and maintains model state for rapid data access.
|
|
/// - `model_card`:
|
|
/// Manages model deployment cards containing detailed metadata, configuration
|
|
/// settings, and versioning information to ensure consistent deployments.
|
|
/// - `preprocessor`:
|
|
/// Provides utilities for transforming raw LLM requests—including tokenization,
|
|
/// prompt formatting, and validation—into a format required by the Dynamo runtime.
|
|
///
|
|
/// Each sub-module is designed to encapsulate its functionality for clean
|
|
/// integration between Python tools and the Dynamo runtime.
|
|
use super::*;
|
|
|
|
pub mod backend;
|
|
pub mod disagg_router;
|
|
pub mod entrypoint;
|
|
pub mod kv;
|
|
pub mod local_model;
|
|
pub mod model_card;
|
|
pub mod preprocessor;
|
|
|
|
#[cfg(feature = "block-manager")]
|
|
pub mod block_manager;
|