Add comprehensive Kronos architecture analysis and documentation - #1

This PR provides a complete analysis of the Kronos codebase architecture in response to the request "分析一下这代码架构" (analyze this code architecture).

What's Added

Comprehensive Documentation

• ARCHITECTURE_ANALYSIS.md - Detailed Chinese analysis covering the two-stage framework, core components, and technical innovations
• TECHNICAL_ANALYSIS.md - In-depth English technical documentation with mathematical foundations and implementation details
• README_ARCHITECTURE.md - Master overview document with usage guidance and best practices

Visual Architecture Diagrams

• figures/architecture_diagram_en.png - English architecture flow diagram showing the complete pipeline
• figures/architecture_diagram.png - Chinese version of the architecture diagram

Key Analysis Findings

Two-Stage Framework Architecture

Kronos employs an innovative two-stage approach:

1. KronosTokenizer - Specialized tokenizer using Binary Spherical Quantization (BSQ) to convert continuous OHLCV data into hierarchical discrete tokens
2. Kronos Model - Autoregressive Transformer that processes quantized tokens for financial forecasting

Core Technical Innovations

• Hierarchical Token System: Dual-channel representation with s1_bits (primary) and s2_bits (auxiliary) tokens
• Dependency-Aware Layer: Cross-attention mechanism modeling conditional dependencies between token types
• Financial Domain Specialization: Purpose-built components for handling high-noise financial time series

Model Variants and Capabilities

Complete End-to-End Solution

The analysis covers the full pipeline including:

• Data preprocessing with Qlib integration
• Multi-stage training (tokenizer → predictor fine-tuning)
• Web interface for demonstrations
• Backtesting and strategy validation tools

Technical Deep Dive

The documentation provides detailed analysis of:

• Binary Spherical Quantization (BSQ): Mathematical foundation and differentiable entropy computation
• Rotary Position Encoding (RoPE): Enhanced temporal relationship modeling
• Conditional Generation: How s2 tokens depend on s1 tokens for improved prediction quality
• Production Deployment: Performance optimization and scaling considerations

This comprehensive analysis serves as both a technical reference and practical guide for understanding, using, and extending the Kronos financial foundation model.

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.