Unused information in token probability distribution of generative LLM: improving LLM reading comprehension through calculation of expected values

2024/06/11 by Krystian Zawistowski, Zawistowski, Krystian · 2 citations
Computer Science · #Artificial Intelligence (cs.AI) #Computation and Language (cs.CL) #FOS: Computer and information sciences #Natural Language Processing Techniques #Text Readability and Simplification

paper · pdf · doi:10.48550/arxiv.2406.10267

openalex publication_date 2024/06/11 · openalex created_date 2024/06/19 · openalex updated_date 2026/07/28

Abstract

LLM text decoding is key component for perceived LLM quality. We demonstrate two experiments showing that decoding methods could be improved by manipulation of token probabilities. First, we test few LLM on SummEval summary scoring dataset, to measure reading comprehension. We compare scores from greedy decoding to expected values over the next token distribution. We scale logits by large temperature to increase the entropy of scores. This allows strong improvement of performance on SummEval (in terms of correlations to human judgement). We see improvement from 6-8% to 13-28% for 7B Mistral and from 20%-46% to 37%-56% for Mixtral, beating GPT 4 0314 result on two metrics. Part of the gain seems related to positional bias. Secondly, we use probability-based tree sampling algorithm, to examine all most probable generations for given prompt.

Cited by

• EntroPIC: Towards Stable Long-Term Training of LLMs via Entropy Stabilization with Proportional-Integral Control
• Explainability-Based Token Replacement on LLM-Generated Text

Related

• Beyond the Black Box: A Statistical Model for LLM Reasoning and Inference
• PLD+: Accelerating LLM inference by leveraging Language Model Artifacts
• Improving LLM-as-a-Judge Inference with the Judgment Distribution
• Explaining and Improving Contrastive Decoding by Extrapolating the Probabilities of a Huge and Hypothetical LM
• On Next-Token Prediction in LLMs: How End Goals Determine the Consistency of Decoding Algorithms