RESEARCH · RESEARCH · #3393
Multi-Future Drafting (MFD) paper proposes mixture block drafters to eliminate chimeras and speed tree-verified speculative decoding
The arXiv preprint (arXiv:2611.13017v1) identifies a misallocation problem in product-ranked tree verification for block drafters (which creates 'chimeras'—prefixes stitched from different futures) and proposes Multi-Future Drafting (MFD). MFD feeds K learned mode embeddings to draft K coherent blocks in one pass and uses a mixture DDTree verifier that can pick the correct future losslessly (with K+1 extra vectors); on four Qwen targets (4B–27B) at batch size 1 MFD reports up to 7.8× speedup and modest percentage gains (5–8% over the best single-mode tree and 9–16% over released DARTree on Qwen3-4B/8B).
KEY POINTS
- The arXiv preprint (arXiv:2611.13017v1) identifies a misallocation problem in product-ranked tree verification for block drafters (which creates 'chimeras'—prefixes stitched from different futures) and proposes Multi-Future Drafting (MFD).
- MFD feeds K learned mode embeddings to draft K coherent blocks in one pass and uses a mixture DDTree verifier that can pick the correct future losslessly (with K+1 extra vectors); on four Qwen targets (4B–27B) at batch size 1 MFD reports up to 7.8× speedup and modest percentage gains (5–8% over the best single-mode tree and 9–16% over released DARTree on Qwen3-4B/8B).
- Fixing the product-ranked-tree misallocation and enabling coherent multi-future drafting can reduce spliced, incorrect prefixes and materially speed up/speculative decoding for large language models.
WHY IT MATTERS
Fixing the product-ranked-tree misallocation and enabling coherent multi-future drafting can reduce spliced, incorrect prefixes and materially speed up/speculative decoding for large language models.