Phoenix Documentation
Phoenix is a high-performance KV cache system designed for efficient memory management in large language model inference. It provides intelligent caching strategies, memory optimization, and seamless integration with AI frameworks.
Welcome to Phoenix documentation.
Contents:
- Phoenix Documentation
- Architecture
- Install & Quick Start
- Adapters
- Kernel Module —
phxfs - LibPhoenix
- GPU 厂商适配指南 / GPU Vendor Porting Guide
- 1. 总览 | Overview
- 2. 厂商驱动的前置条件 | Prerequisites on the vendor driver stack
- 3. 内核侧:
module/<vendor>-backend.c| Kernel side: the P2P backend - 4. 用户态:
libphoenix/connectors/<vendor>_connector.cpp| User side: the DevConnector - 5. FULL / STAGING 模式对新厂商意味着什么 | FULL vs STAGING for a new vendor
- 6. 验证清单 | Validation checklist
- 7. 相关文档 | Related documents
- Troubleshooting — Kernel Module Installation
- Roadmap
Key Features
Intelligent KV Cache Management: Optimized memory allocation for transformer models
High Performance: Low-latency cache operations for real-time inference
Multi-Backend Support: Works with various GPU architectures
Framework Integration: Easy integration with popular AI frameworks
Links
GitHub Repository: https://github.com/xpu-io/Phoenix
License: See LICENSE file in the repository