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, thereby maintaining output quality. 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 fine-tuned models demonstrated a speedup up to 1.73 x.
With all the interest around speculative decoding using a smaller model, this presents an interesting opportunity to speed up without needing the extra space for a draft model
Definitely very interesting, but figuring out what layers to skip is a relatively difficult problem.
I really wish they’d shown an example of the optimal layers to skip for the 13B model. Like the paper notes, using the wrong combination of skipped layers can be worse overall. So it’s not just about how many layers you skip, but which ones as well.
It would also be interesting to see if there are any common patterns in which layers are most skippable. It probably would be model architecture specific but it would be pretty useful if you could calculate the optimal skip pattern for say a 3B model and then translate that to a 30B with good/reasonable results.
The good news is if you do it wrong, much like regular speculative generation, you will still get the right result that the full model would output at the end, so there won’t be any loss in quality, just loss in speed
It’s definitely a good point tho, finding the optimal configuration is the difference between slower/minimal speedup and potentially huge amounts of speedup
