Author name cluster

Sumit Sanghai

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

4 papers

2 author rows

ICLR Conference 2024 Conference Paper

Functional Interpolation for Relative Positions improves Long Context Transformers

Shanda Li
Chong You
Guru Guruganesh
Joshua Ainslie
Santiago Ontañón
Manzil Zaheer
Sumit Sanghai
Yiming Yang 0002

Preventing the performance decay of Transformers on inputs longer than those used for training has been an important challenge in extending the context length of these models. Though the Transformer architecture has fundamentally no limits on the input sequence lengths it can process, the choice of position encoding used during training can limit the performance of these models on longer inputs. We propose a novel functional relative position encoding with progressive interpolation, FIRE, to improve Transformer generalization to longer contexts. We theoretically prove that this can represent some of the popular relative position encodings, such as T5's RPE, Alibi, and Kerple. We next empirically show that FIRE models have better generalization to longer contexts on both zero-shot language modeling and long text benchmarks.

Details

ICML Conference 2023 Conference Paper

Arithmetic Sampling: Parallel Diverse Decoding for Large Language Models

Luke Vilnis
Yury Zemlyanskiy
Patrick Murray
Alexandre Passos
Sumit Sanghai

Decoding methods for large language models often trade-off between diversity of outputs and parallelism of computation. Methods such as beam search and Gumbel top-k sampling can guarantee a different output for each element of the beam, but are not easy to parallelize. Alternatively, methods such as temperature sampling and its modifications (top-k sampling, nucleus sampling, typical decoding, and others), are embarrassingly parallel, but have no guarantees about duplicate samples. We present a framework for sampling according to an arithmetic code book implicitly defined by a large language model, compatible with common sampling variations, with provable beam diversity under certain conditions, as well as being embarrassingly parallel and providing unbiased and consistent expectations from the original model. We demonstrate the effectiveness of our approach on WMT machine translation, more than halving the standard deviation when estimating expected BLEU score reward, and closing the BLEU score gap between independent sampling and beam search by up to 63%.

Details

ICML Conference 2023 Conference Paper

Pre-computed memory or on-the-fly encoding? A hybrid approach to retrieval augmentation makes the most of your compute

Michiel de Jong
Yury Zemlyanskiy
Nicholas FitzGerald
Joshua Ainslie
Sumit Sanghai
Fei Sha
William W. Cohen

Retrieval-augmented language models such as Fusion-in-Decoder are powerful, setting the state of the art on a variety of knowledge-intensive tasks. However, they are also expensive, due to the need to encode a large number of retrieved passages. Some work avoids this cost by pre-encoding a text corpus into a memory and retrieving dense representations directly. However, pre-encoding memory incurs a severe quality penalty as the memory representations are not conditioned on the current input. We propose LUMEN, a hybrid between these two extremes, pre-computing the majority of the retrieval representation and completing the encoding on the fly using a live encoder that is conditioned on the question and fine-tuned for the task. We show that LUMEN significantly outperforms pure memory on multiple question-answering tasks while being much cheaper than FiD, and outperforms both for any given compute budget. Moreover, the advantage of LUMEN over FiD increases with model size.

Details

IJCAI Conference 2003 Conference Paper

Dynamic Probabilistic Relational Models

Sumit Sanghai
Pedro Domingos
Daniel Weld

Intelligent agents must function in an uncertain world, containing multiple objects and relations that change over time. Unfortunately, no representation is currently available that can handle all these issues, while allowing for principled and efficient inference. This paper addresses this need by introducing dynamic probabilistic relational models (DPRMs). DPRMs are an extension of dynamic Bayesian networks (DBNs) where each time slice (and its dependences on previous slices) is represented by a probabilistic relational model (PRM). Particle filtering, the standard method for inference in DBNs, has severe limitations when applied to DPRMs, but we are able to greatly improve its performance through a form of relational Rao-Blackwellisation. Further gains in efficiency arc obtained through the use of abstraction trees, a novel data structure. We successfully apply DPRMs to execution monitoring and fault diagnosis of an assembly plan, in which a complex product is gradually constructed from subparts.

PDF