Evaluation of Large Language Models via Coupled Token Generation
- URL: http://arxiv.org/abs/2502.01754v1
- Date: Mon, 03 Feb 2025 19:01:17 GMT
- Title: Evaluation of Large Language Models via Coupled Token Generation
- Authors: Nina Corvelo Benz, Stratis Tsirtsis, Eleni Straitouri, Ivi Chatzi, Ander Artola Velasco, Suhas Thejaswi, Manuel Gomez-Rodriguez,
- Abstract summary: State of the art large language models rely on randomization to respond to a prompt.
We argue that the evaluation and ranking of large language models should control for the randomization underpinning their functioning.
- Score: 15.0803535133762
- License:
- Abstract: State of the art large language models rely on randomization to respond to a prompt. As an immediate consequence, a model may respond differently to the same prompt if asked multiple times. In this work, we argue that the evaluation and ranking of large language models should control for the randomization underpinning their functioning. Our starting point is the development of a causal model for coupled autoregressive generation, which allows different large language models to sample responses with the same source of randomness. Building upon our causal model, we first show that, on evaluations based on benchmark datasets, coupled autoregressive generation leads to the same conclusions as vanilla autoregressive generation but using provably fewer samples. However, we further show that, on evaluations based on (human) pairwise comparisons, coupled and vanilla autoregressive generation can surprisingly lead to different rankings when comparing more than two models, even with an infinite amount of samples. This suggests that the apparent advantage of a model over others in existing evaluation protocols may not be genuine but rather confounded by the randomness inherent to the generation process. To illustrate and complement our theoretical results, we conduct experiments with several large language models from the Llama family. We find that, across multiple knowledge areas from the popular MMLU benchmark dataset, coupled autoregressive generation requires up to 40% fewer samples to reach the same conclusions as vanilla autoregressive generation. Further, using data from the LMSYS Chatbot Arena platform, we find that the win-rates derived from pairwise comparisons by a strong large language model to prompts differ under coupled and vanilla autoregressive generation.
Related papers
- Promises and Pitfalls of Generative Masked Language Modeling: Theoretical Framework and Practical Guidelines [74.42485647685272]
We focus on Generative Masked Language Models (GMLMs)
We train a model to fit conditional probabilities of the data distribution via masking, which are subsequently used as inputs to a Markov Chain to draw samples from the model.
We adapt the T5 model for iteratively-refined parallel decoding, achieving 2-3x speedup in machine translation with minimal sacrifice in quality.
arXiv Detail & Related papers (2024-07-22T18:00:00Z) - mFACE: Multilingual Summarization with Factual Consistency Evaluation [79.60172087719356]
Abstractive summarization has enjoyed renewed interest in recent years, thanks to pre-trained language models and the availability of large-scale datasets.
Despite promising results, current models still suffer from generating factually inconsistent summaries.
We leverage factual consistency evaluation models to improve multilingual summarization.
arXiv Detail & Related papers (2022-12-20T19:52:41Z) - Robustness Analysis of Deep Learning Models for Population Synthesis [5.9106199000537645]
We present bootstrap confidence interval for the deep generative models to evaluate robustness to multiple datasets.
The models are implemented on multiple travel diaries of Montreal Origin- Destination Survey of 2008, 2013, and 2018.
Results show that the predictive errors of CTGAN have narrower confidence intervals indicating its robustness to multiple datasets.
arXiv Detail & Related papers (2022-11-23T22:55:55Z) - Twist Decoding: Diverse Generators Guide Each Other [116.20780037268801]
We introduce Twist decoding, a simple and general inference algorithm that generates text while benefiting from diverse models.
Our method does not assume the vocabulary, tokenization or even generation order is shared.
arXiv Detail & Related papers (2022-05-19T01:27:53Z) - Comparing hundreds of machine learning classifiers and discrete choice models in predicting travel behavior: an empirical benchmark [6.815730801645785]
Many studies have compared machine learning (ML) and discrete choice models (DCMs) in predicting travel demand.
These studies often lack generalizability as they compare models deterministically without considering contextual variations.
This benchmark study compares two large-scale data sources.
arXiv Detail & Related papers (2021-02-01T19:45:47Z) - Improving the Reconstruction of Disentangled Representation Learners via Multi-Stage Modeling [54.94763543386523]
Current autoencoder-based disentangled representation learning methods achieve disentanglement by penalizing the ( aggregate) posterior to encourage statistical independence of the latent factors.
We present a novel multi-stage modeling approach where the disentangled factors are first learned using a penalty-based disentangled representation learning method.
Then, the low-quality reconstruction is improved with another deep generative model that is trained to model the missing correlated latent variables.
arXiv Detail & Related papers (2020-10-25T18:51:15Z) - Multi-Referenced Training for Dialogue Response Generation [36.24321477524634]
We show that gap between the real world probability distribution and the single-referenced data's probability distribution prevents the model from learning the one-to-many relations efficiently.
We generate diverse pseudo references from a powerful pretrained model to build multi-referenced data that provides a better approximation of the real-world distribution.
arXiv Detail & Related papers (2020-09-15T14:17:53Z) - On the Discrepancy between Density Estimation and Sequence Generation [92.70116082182076]
log-likelihood is highly correlated with BLEU when we consider models within the same family.
We observe no correlation between rankings of models across different families.
arXiv Detail & Related papers (2020-02-17T20:13:35Z) - Learning to Compare for Better Training and Evaluation of Open Domain
Natural Language Generation Models [23.62054164511058]
We propose to evaluate natural language generation models by learning to compare a pair of generated sentences by fine-tuning BERT.
While able to be trained in a fully self-supervised fashion, our model can be further fine-tuned with a little amount of human preference annotation.
arXiv Detail & Related papers (2020-02-12T15:52:21Z) - AvgOut: A Simple Output-Probability Measure to Eliminate Dull Responses [97.50616524350123]
We build dialogue models that are dynamically aware of what utterances or tokens are dull without any feature-engineering.
The first model, MinAvgOut, directly maximizes the diversity score through the output distributions of each batch.
The second model, Label Fine-Tuning (LFT), prepends to the source sequence a label continuously scaled by the diversity score to control the diversity level.
The third model, RL, adopts Reinforcement Learning and treats the diversity score as a reward signal.
arXiv Detail & Related papers (2020-01-15T18:32:06Z)
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.