SCX Router: Streaming Zero-Shot Model Selection with a Decoder-KV Classifier and a Real-World Task Ontology
A lightweight GLiClass-based router is introduced, a lightweight GLiClass-based router that assigns a suitability score to each inference-time model label without autoregressive generation, and a released 0.6B-parameter checkpoint combines a Qwen3 decoder with a shallow bidirectional scorer.
Ihor Stepanov, Aleksandr Smechov, Mykhailo Shtopko et al.
· 0 citations