DistillSpec: Improving Speculative Decoding via Knowledge Distillation

• URL: http://arxiv.org/abs/2310.08461v2
• Date: Sun, 31 Mar 2024 03:06:51 GMT
• Title: DistillSpec: Improving Speculative Decoding via Knowledge Distillation
• Authors: Yongchao Zhou, Kaifeng Lyu, Ankit Singh Rawat, Aditya Krishna Menon, Afshin Rostamizadeh, Sanjiv Kumar, Jean-François Kagy, Rishabh Agarwal,
• Abstract summary: Speculative decoding (SD) accelerates large language model inference by employing a faster draft model for generating multiple tokens. We propose DistillSpec that uses knowledge distillation to better align the draft model with the target model, before applying SD. We show that DistillSpec yields impressive 10 - 45% speedups over standard SD on a range of standard benchmarks.
• Score: 70.61777015900272
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Speculative decoding (SD) accelerates large language model inference by employing a faster draft model for generating multiple tokens, which are then verified in parallel by the larger target model, resulting in the text generated according to the target model distribution. However, identifying a compact draft model that is well-aligned with the target model is challenging. To tackle this issue, we propose DistillSpec that uses knowledge distillation to better align the draft model with the target model, before applying SD. DistillSpec makes two key design choices, which we demonstrate via systematic study to be crucial to improving the draft and target alignment: utilizing on-policy data generation from the draft model, and tailoring the divergence function to the task and decoding strategy. Notably, DistillSpec yields impressive 10 - 45% speedups over standard SD on a range of standard benchmarks, using both greedy and non-greedy sampling. Furthermore, we combine DistillSpec with lossy SD to achieve fine-grained control over the latency vs. task performance trade-off. Finally, in practical scenarios with models of varying sizes, first using distillation to boost the performance of the target model and then applying DistillSpec to train a well-aligned draft model can reduce decoding latency by 6-10x with minimal performance drop, compared to standard decoding without distillation.

Related papers

• ParallelSpec: Parallel Drafter for Efficient Speculative Decoding [62.68430939686566]
  We present ParallelSpec, an alternative to auto-regressive drafting strategies in state-of-the-art speculative decoding approaches. In contrast to auto-regressive drafting in the speculative stage, we train a parallel drafter to serve as an efficient speculative model.
  arXiv  Detail & Related papers  (2024-10-08T01:05:08Z)
• Improving Multi-candidate Speculative Decoding [1.6291177798903276]
  Speculative Decoding (SD) is a technique to accelerate the inference of Large Language Models (LLMs) In this work, we introduce a new version of MCSD that includes a target model multi-candidate generation. We also evaluate the effects of using the target model multi-candidate process with different draft models on output quality.
  arXiv  Detail & Related papers  (2024-09-16T18:20:38Z)
• Graph-Structured Speculative Decoding [52.94367724136063]
  Speculative decoding has emerged as a promising technique to accelerate the inference of Large Language Models. We introduce an innovative approach utilizing a directed acyclic graph (DAG) to manage the drafted hypotheses. We observe a remarkable speedup of 1.73$times$ to 1.96$times$, significantly surpassing standard speculative decoding.
  arXiv  Detail & Related papers  (2024-07-23T06:21:24Z)
• S2D: Sorted Speculative Decoding For More Efficient Deployment of Nested Large Language Models [32.68002253527712]
  We introduce a novel multi-target scenario for the deployment of draft models for faster inference. We present a novel, more efficient sorted speculative decoding mechanism that outperforms regular baselines in multi-target settings.
  arXiv  Detail & Related papers  (2024-07-02T05:14:15Z)
• Direct Alignment of Draft Model for Speculative Decoding with Chat-Fine-Tuned LLMs [11.245862832561176]
  Training a high-quality draft model is required to enable inference acceleration via speculative decoding. We train Llama 2 Chat Drafter 115M, a draft model for Llama 2 Chat 7B or larger, with only 1.64% of the original size. Our results show that Llama 2 Chat Drafter 115M with speculative decoding achieves up to 2.3 block efficiency and 2.4$times$ speed-up.
  arXiv  Detail & Related papers  (2024-02-29T19:55:06Z)
• Multi-Candidate Speculative Decoding [82.05519287513444]
  Large language models have shown impressive capabilities across a variety of NLP tasks, yet their generating text autoregressively is time-consuming. One way to speed them up is speculative decoding, which generates candidate segments from a fast draft model that is then verified in parallel by the target model. This paper proposes sampling multiple candidates from a draft model and then organising them in batches for verification. We design algorithms for efficient multi-candidate verification while maintaining the distribution of the target model.
  arXiv  Detail & Related papers  (2024-01-12T17:15:23Z)
• Cascade Speculative Drafting for Even Faster LLM Inference [25.642604897018852]
  Speculative decoding improves the efficiency of large language model (LLM) inference. We introduce Cascade Speculative Drafting (CS Drafting), a speculative execution algorithm that incorporates two types of cascades. CS Drafting achieves up to an 81 percent additional speedup over speculative decoding in our experiments.
  arXiv  Detail & Related papers  (2023-12-18T18:59:46Z)
• Online Speculative Decoding [34.987825705622555]
  We introduce online speculative decoding to accelerate the inference of large language models. The main idea is to continuously update the (multiple) draft model(s) on observed user query data. We develop a prototype of online speculative decoding based on knowledge distillation and evaluate it using both synthetic and real query data.
  arXiv  Detail & Related papers  (2023-10-11T04:03:42Z)
• Dual Student Networks for Data-Free Model Stealing [79.67498803845059]
  Two main challenges are estimating gradients of the target model without access to its parameters, and generating a diverse set of training samples. We propose a Dual Student method where two students are symmetrically trained in order to provide the generator a criterion to generate samples that the two students disagree on. We show that our new optimization framework provides more accurate gradient estimation of the target model and better accuracies on benchmark classification datasets.
  arXiv  Detail & Related papers  (2023-09-18T18:11:31Z)
• Generic-to-Specific Distillation of Masked Autoencoders [119.21281960831651]
  We propose generic-to-specific distillation (G2SD) to tap the potential of small ViT models under the supervision of large models pre-trained by masked autoencoders. With G2SD, the vanilla ViT-Small model achieves 98.7%, 98.1% and 99.3% the performance of its teacher for image classification, object detection, and semantic segmentation.
  arXiv  Detail & Related papers  (2023-02-28T17:13:14Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.