Accurate multi-step traffic flow forecasting requires dynamic spatial modeling and adaptation to heterogeneous temporal and node-level patterns. This study proposes FAMoE-ST, a hierarchically frozen attention network with a hybrid-memory mixture of experts. Historical flow and temporal and node context are encoded as sensor tokens and processed by a six-layer Transformer initialized from GPT-2. To remove dependence on arbitrary sensor indexing, causal masking is replaced by all-to-all bidirectional spatial attention, and all nodes share the zero GPT position ID. The lower four blocks are frozen, whereas the upper two blocks are adapted and their feed-forward networks are replaced by four-expert modules. A linear router is fused with a 32-slot memory router; each token retrieves four slots and activates two experts. With 12 observations predicting the next 12 steps, FAMoE-ST achieves MAE/RMSE/MAPE values of 18.55/30.35/12.91% on PEMS04 and 14.57/24.08/9.71% on PEMS08. Relative to ST-LLM, MAE decreases by 6.97% and 7.39%, respectively. Ablations show complementary benefits from restricted adaptation, sparse experts, and memory routing. A parameter-matched control further indicates that the gains are not attributable solely to stored capacity, although the experiments do not isolate the effect of GPT-2 pretraining.