Uncovering stealthy Distributed Denial of Service (DDoS) attacks at the network-edge remains challenging due to increased system complexity and users' access to digital services. Therefore, in this work, we develop TinySEED, a lightweight yet accurate DDoS detection model based on TinyBERT that integrates a channel attention mechanism. TinyBERT is developed via knowledge distillation from BERT, providing a compact backbone that preserves key representational capacity while significantly reducing model size and inference latency. To further enhance discriminative power between DDoS attacks and benign traces, we integrate channel attention, which adaptively re-weights embedding channels (dimensions), enabling the model to capture fine-grained distinctions between benign and malicious flows that are often overlooked by distilled transformers. Extensive DDoS detection experiments across five benchmark datasets demonstrate that TinySEED consistently outperforms conventional machine learning methods, neural architectures, and transformer-based baselines. TinySEED achieves higher detection accuracy on benchmarks while maintaining low inference latency. These findings highlight the effectiveness of combining efficient transformer distillation with channel attention to achieve a practical and robust solution for low-latency DDoS detection in network-edge environments.