English
← Back to AI Technology

Inference and deployment

Speculative Decoding

An inference method in which a faster draft proposes multiple tokens and the target model verifies them in parallel. It can reduce generation latency without changing the target distribution.

Reference period
2022
Last reviewed

Key terms

  • draft model
  • verification
  • latency
  • exact sampling

Connected across the project

Primary sources

  1. Fast Inference from Transformers via Speculative Decoding

This is a curated technical map, not a claim of comprehensive coverage.