Methods › General › Parameter Sharing › Weight Tying

Weight Tying

63 papers tagged archive 2025-07-28

Introduced by Ofir Press et al. in Using the Output Embedding to Improve Language Models

archive 2025-07-28 Description, source and code snippet are the archive's method entry.

Weight Tying improves the performance of language models by tying (sharing) the weights of the embedding and softmax layers. This method also massively reduces the total number of parameters in the language models that it is applied to.

Language models are typically comprised of an embedding layer, followed by a number of Transformer or LSTM layers, which are finally followed by a softmax layer. Embedding layers learn word representations, such that similar words (in meaning) are represented by vectors that are near each other (in cosine distance). [Press & Wolf, 2016] showed that the softmax matrix, in which every word also has a vector representation, also exhibits this property. This leads them to propose to share the softmax and embedding matrices, which is done today in nearly all language models.

This method was independently introduced by Press & Wolf, 2016 and Inan et al, 2016.

Additionally, the Press & Wolf paper proposes Three-way Weight Tying, a method for NMT models in which the embedding matrix for the source language, the embedding matrix for the target language, and the softmax matrix for the target language are all tied. That method has been adopted by the Attention Is All You Need model and many other neural machine translation models.

PaperSource

Papers archive 2025-07-28

30 shown of 63, newest first. Repository counts are the archive's code-links table. A Syntology line states what Syntology ran from that paper's harvested code; it is per sample and not a correctness claim.

Tasks archive 2025-07-28

20 shown of 76 tasks the archive attaches to papers tagged with this method, by distinct papers. A task without a page in the catalog is plain text.

TaskPapers
Language Modelling23
Language Modeling21
Transfer Learning16
General Classification15
Text Classification15
text-classification11
Sentiment Analysis9
Classification8
Translation7
Machine Translation5
Word Embeddings5
Language Identification4
Decision Making3
Hate Speech Detection3
Sentence3
Sentiment Classification3
Articles2
BIG-bench Machine Learning2
Decoder2
Image Classification2

Usage over time archive 2025-07-28

Papers per year tagged with Weight Tying: 2016 to 2025, peak 16 16 0 2016: 2 papers 2016 2017: 3 papers 2017 2018: 4 papers 2018 2019: 16 papers 2019 2020: 15 papers 2020 2021: 13 papers 2021 2022: 2 papers 2022 2023: 5 papers 2023 2024: 2 papers 2024 2025: 1 paper 2025
Papers per year the archive tags with this method, by the paper's archive date (63 dated). Bars are counts, not a trend claim.

Components: the archive holds no method-to-method composition, so PwC's Components table cannot be rebuilt; the Papers list carries no Results column for the same reason (the archive does not join its leaderboard rows to method tags).

Categories archive 2025-07-28

Parameter Sharing

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections