金融K线数据基础模型微调指南生成器

You are a quantitative researcher specializing in financial foundation models and time series analysis. Help me create a comprehensive fine-tuning guide for a financial candlestick (K-line) foundation model.

Based on my use case, generate:

1. Data Preparation Pipeline:

   • Required data format (OHLCV fields, timeframe alignment)
   • Data cleaning: handle missing candles, outlier detection, corporate actions
   • Train/validation/test split strategy (walk-forward, expanding window)
   • Tokenization approach for continuous financial data

2. Fine-tuning Strategy:

   • Full fine-tuning vs LoRA vs prefix tuning — recommend based on data size
   • Learning rate schedule (warmup + cosine decay)
   • Batch size and gradient accumulation settings
   • Regularization: dropout, weight decay, early stopping criteria

3. Task-Specific Heads:

   • Price direction classification
   • Volatility regime detection
   • Support/resistance level prediction
   • Multi-step price forecasting

4. Evaluation Framework:

   • Financial metrics: Sharpe ratio, max drawdown, hit rate
   • ML metrics: MSE, MAE, directional accuracy
   • Backtesting protocol with transaction costs

5. Production Deployment:

   • Model serving latency requirements
   • Real-time inference pipeline design
   • Model monitoring and drift detection

My use case: [describe your target market, asset class, and trading frequency]