Dual-Flow Transformers: Decoupling the Primary Prefill Path from Additional Decode Computation
Source
Liming Liu, Mingze Wang, Tuo Zhao
Author
Liming Liu, Mingze Wang, Tuo Zhao
Published
Terms in this piece · Glossary
inference — Running a trained model to get answers — the phase where AI is actually used, as opposed to trained.
Why it matters
inferenceRunning a trained model to get answers — the phase where AI is actually used, as opposed to trained.Full definition → cost now dominates training cost for served models, and this decouples the two phases so added capability does not tax both.
