Draft & Verify: Lossless Large Language Model Acceleration via Self-Speculative Decoding
This provides a plug-and-play, cost-effective solution for accelerating LLM inference, which is incremental as it builds on speculative decoding but eliminates the need for training or extra memory.
The paper tackles the problem of slow inference in Large Language Models by introducing self-speculative decoding, a method that accelerates generation without auxiliary models, achieving up to 1.99x speedup on benchmarks with LLaMA-2 variants while ensuring lossless output.
We present a novel inference scheme, self-speculative decoding, for accelerating Large Language Models (LLMs) without the need for an auxiliary model. This approach is characterized by a two-stage process: drafting and verification. The drafting stage generates draft tokens at a slightly lower quality but more quickly, which is achieved by selectively skipping certain intermediate layers during drafting. Subsequently, the verification stage employs the original LLM to validate those draft output tokens in one forward pass. This process ensures the final output remains identical to that produced by the unaltered LLM. Moreover, the proposed method requires no additional neural network training and no extra memory footprint, making it a plug-and-play and cost-effective solution for inference acceleration. Benchmarks with LLaMA-2 and its variants demonstrated a speedup up to 1.99$\times$.