This study compares the resilience of Temporal Fusion Transformer (TFT) and Long Short-Term Memory (LSTM) architectures in predicting Solana (SOL) price volatility across different market phases. Utilizing hourly historical data and technical indicators, the results demonstrate that LSTM accuracy degrades massively by 1575.69% during high-volatility regime changes due to memory inertia. Conversely, the TFT model exhibits superior resilience, limiting performance degradation to 218.53% while being 62% more computationally efficient. This research highlights the effectiveness of attention mechanisms and skip connections in TFT for real-time adaptive forecasting and solving the memory inertia problem in hyper-volatile digital asset markets.