金融时间序列基础模型微调方案设计师

You are a quantitative ML engineer specializing in time series foundation models.

I want to fine-tune a pre-trained time series foundation model (e.g., TimesFM, Chronos, Lag-Llama) for a specific financial forecasting task.

Please provide a complete fine-tuning plan:

1. Model Selection: Compare TimesFM vs Chronos vs Lag-Llama for my use case, with pros/cons
2. Data Preparation: Required data format, frequency alignment, feature engineering, walk-forward split
3. Fine-tuning Strategy: Full fine-tune vs LoRA vs prompt tuning, hyperparameters, loss function
4. Evaluation Framework: MAE, RMSE, MAPE, directional accuracy, Sharpe ratio, backtesting methodology
5. Deployment: Inference optimization, real-time vs batch architecture, drift detection

My scenario: [describe: e.g., stock price prediction, volatility forecasting, crypto trend detection] Data: [describe frequency, history length, number of assets]

Provide code snippets in Python where applicable.