Double: Breaking the Acceleration Limit via Double Retrieval - a Piggy12 Collection

Piggy12 's Collections

Double: Breaking the Acceleration Limit via Double Retrieval

Speculative Decoding via Hybrid Drafting and Rollback-Aware

Double: Breaking the Acceleration Limit via Double Retrieval

updated Feb 21

Double: Breaking the Acceleration Limit via Double Retrieval Speculative Parallelism

Paper • 2601.05524 • Published Jan 9 • 1