Greedy Decoding

فك الترميز الجشع

استراتيجية توليد يختار فيها النموذج الرمز الأعلى احتمالاً في كل خطوة زمنية دون استكشاف بدائل، وهي الطريقة المستخدمة في تجارب سلسلة التفكير الأصلية.

A generation strategy where the model picks the highest-probability token at each time step without exploring alternatives, the method used in the original chain-of-thought experiments.

Also translated asالاختيار الأعظمي عند كل خطوة، الاختيار الفوري للكلمة الأعلى احتمالاً، التوليد الجشع، التوليد النصي الحتمي المباشر

First appears in this corpus in: Sequence to Sequence Learning with Neural Networks (2014)

Appears in these papers

Fusion-in-Decoder: Leveraging Passage Retrieval with Generative Models for Open Domain Question Answering2021 in the sky ✦
Fusion-in-Decoder: Leveraging Passage Retrieval with Generative Models for Open Domain Question Answering2021 in the sky ✦
Learning to Summarize from Human Feedback2020 in the sky ✦
Self-Consistency Improves Chain of Thought Reasoning in Language Models2022 in the sky ✦
Self-Consistency Improves Chain of Thought Reasoning in Language Models2022 in the sky ✦
Sequence to Sequence Learning with Neural Networks2014 in the sky ✦
Show and Tell: A Neural Image Caption Generator2015 in the sky ✦
Fast Inference from Transformers via Speculative Decoding2023 in the sky ✦
Robust Speech Recognition via Large-Scale Weak Supervision2022 in the sky ✦