SpecExit Accelerates Large Reasoning Models with Early Exit

SpecExit cuts generation length by roughly 66 % and achieves about a 2.5× speedup in latency for large reasoning models, while keeping accuracy stable. getnews.me/specexit-accelerates-lar... #specexit #largereasoning #earlyexit