Superposed Decoding: Multiple Generations from a Single Autoregressive Inference Pass

• URL: http://arxiv.org/abs/2405.18400v6
• Date: Wed, 30 Oct 2024 21:22:54 GMT
• Title: Superposed Decoding: Multiple Generations from a Single Autoregressive Inference Pass
• Authors: Ethan Shen, Alan Fan, Sarah M. Pratt, Jae Sung Park, Matthew Wallingford, Sham M. Kakade, Ari Holtzman, Ranjay Krishna, Ali Farhadi, Aditya Kusupati,
• Abstract summary: Superposed Decoding is a new decoding algorithm that generates $k$ drafts at the cost of one autoregressive inference pass. Superposed Decoding can be combined with other decoding strategies, resulting in universal coverage gains when scaling inference time compute.
• Score: 72.07642648108849
• License:
• Abstract: Many applications today provide users with multiple auto-complete drafts as they type, including GitHub's code completion, Gmail's smart compose, and Apple's messaging auto-suggestions. Under the hood, language models support this by running an autoregressive inference pass to provide a draft. Consequently, providing $k$ drafts to the user requires running an expensive language model $k$ times. To alleviate the computation cost of running $k$ inference passes, we propose Superposed Decoding, a new decoding algorithm that generates $k$ drafts at the computation cost of one autoregressive inference pass. We achieve this by feeding a superposition of the most recent token embeddings from the $k$ drafts as input to the next decoding step of the language model. At every inference step we combine the $k$ drafts with the top-$k$ tokens to get $k^2$ new drafts and cache the $k$ most likely options, using an n-gram interpolation with minimal compute overhead to filter out incoherent generations. Our experiments show that $k$ drafts from Superposed Decoding are at least as coherent and factual as Nucleus Sampling and Greedy Decoding respectively, while being at least $2.44\times$ faster for $k\ge3$. In a compute-normalized setting, user evaluations demonstrably favor text generated by Superposed Decoding over Nucleus Sampling. Superposed Decoding can also be combined with other decoding strategies, resulting in universal coverage gains when scaling inference time compute. Code and more examples open-sourced at https://github.com/RAIVNLab/SuperposedDecoding.

Related papers

• SuffixDecoding: A Model-Free Approach to Speeding Up Large Language Model Inference [9.143856130336783]
  SuffixDecoding is a model-free approach to accelerating large language model (LLM) inference through speculative decoding. Our approach enables flexible tree-structured speculation without the overhead of maintaining and orchestrating additional models. For a proprietary multi-LLM text-to-token application, SuffixDecoding achieves up to $2.9times$ higher output throughput and $3times$ lower latency than speculative decoding.
  arXiv  Detail & Related papers  (2024-11-07T18:49:33Z)
• Some Notes on the Sample Complexity of Approximate Channel Simulation [2.4554686192257424]
  Channel simulation algorithms can efficiently encode random samples from a prescribed target distribution $Q$ and find applications in machine learning-based lossy data compression. This paper considers approximate schemes with a fixed runtime instead. We exploit global-bound, depth-limited A* coding to ensure $mathrmTV[Q Vert P] leq epsilon$ and maintain optimal coding performance with a sample complexity of only $expbig((D_KL[Q Vert P] + o(1)) big/ epsilonbig
  arXiv  Detail & Related papers  (2024-05-07T14:44:41Z)
• Ouroboros: Generating Longer Drafts Phrase by Phrase for Faster Speculative Decoding [65.94521678103237]
  Speculative decoding is a widely used method that accelerates the generation process of large language models. We introduce Ouroboros, which can generate draft phrases to parallelize the drafting process. Ouroboros can achieve speedups of up to $2.8times$ over speculative decoding and $3.9times$ over vanilla decoding.
  arXiv  Detail & Related papers  (2024-02-21T11:31:28Z)
• Triple-Encoders: Representations That Fire Together, Wire Together [51.15206713482718]
  Contrastive Learning is a representation learning method that encodes relative distances between utterances into the embedding space via a bi-encoder. This study introduces triple-encoders, which efficiently compute distributed utterance mixtures from these independently encoded utterances. We find that triple-encoders lead to a substantial improvement over bi-encoders, and even to better zero-shot generalization than single-vector representation models.
  arXiv  Detail & Related papers  (2024-02-19T18:06:02Z)
• Linear-time Minimum Bayes Risk Decoding with Reference Aggregation [52.1701152610258]
  Minimum Bayes Risk (MBR) decoding is a text generation technique that has been shown to improve the quality of machine translations. It requires the pairwise calculation of a utility metric, which has quadratic complexity. We propose to approximate pairwise metric scores with scores calculated against aggregated reference representations.
  arXiv  Detail & Related papers  (2024-02-06T18:59:30Z)
• SpecTr: Fast Speculative Decoding via Optimal Transport [30.18181671899423]
  We develop a new autoregressive sampling algorithm called $textitSpecTr$, which provides speedup in decoding while ensuring that there is no quality degradation in the decoded output. We experimentally demonstrate that for state-of-the-art large language models, the proposed approach achieves a wall clock speedup of 2.13X, a further 1.37X speedup over speculative decoding on standard benchmarks.
  arXiv  Detail & Related papers  (2023-10-23T17:47:34Z)
• Interactive Code Generation via Test-Driven User-Intent Formalization [60.90035204567797]
  Large language models (LLMs) produce code from informal natural language (NL) intent. It is hard to define a notion of correctness since natural language can be ambiguous and lacks a formal semantics. We describe a language-agnostic abstract algorithm and a concrete implementation TiCoder.
  arXiv  Detail & Related papers  (2022-08-11T17:41:08Z)
• Learning Autoencoders with Relational Regularization [89.53065887608088]
  A new framework is proposed for learning autoencoders of data distributions. We minimize the discrepancy between the model and target distributions, with a emphrelational regularization We implement the framework with two scalable algorithms, making it applicable for both probabilistic and deterministic autoencoders.
  arXiv  Detail & Related papers  (2020-02-07T17:27:30Z)

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.