this post was submitted on 18 Jul 2023
5 points (100.0% liked)

Machine Learning

4 readers
1 users here now

Machine learning (ML) is a field devoted to understanding and building methods that let machines "learn" – that is, methods that leverage data to improve computer performance on some set of tasks. Machine learning algorithms build a model based on sample data, known as training data, in order to make predictions or decisions without being explicitly programmed to do so. Machine learning algorithms are used in a wide variety of applications, such as in medicine, email filtering, speech recognition, agriculture, and computer vision, where it is difficult or unfeasible to develop conventional algorithms to perform the needed tasks.

founded 1 year ago
 

This is an exciting new paper that replaces attention in the Transformer architecture with a set of decomposable matrix operations that retain the modeling capacity of Transformer models, while allowing parallel training and efficient RNN-like inference without the use of attention (it doesn't use a softmax).

It achieves lower perplexity than Transformers models with more than 2B parameters and requires much lower GPU memory and FLOPs compared Transformers for inference.

Abstract:

In this work, we propose Retentive Network (RetNet) as a foundation architecture for large language models, simultaneously achieving training parallelism, low-cost inference, and good performance. We theoretically derive the connection between recurrence and attention. Then we propose the retention mechanism for sequence modeling, which supports three computation paradigms, i.e., parallel, recurrent, and chunkwise recurrent. Specifically, the parallel representation allows for training parallelism. The recurrent representation enables low-cost $O(1)$ inference, which improves decoding throughput, latency, and GPU memory without sacrificing performance. The chunkwise recurrent representation facilitates efficient long-sequence modeling with linear complexity, where each chunk is encoded parallelly while recurrently summarizing the chunks. Experimental results on language modeling show that RetNet achieves favorable scaling results, parallel training, low-cost deployment, and efficient inference. The intriguing properties make RetNet a strong successor to Transformer for large language models. Code will be available at https://aka.ms/retnet.

you are viewing a single comment's thread
view the rest of the comments
[–] Lenguador@kbin.social 3 points 1 year ago (1 children)

This looks amazing, if true. The paper is claiming state of the art across literally every metric. Even in their ablation study the model outperforms all others.

I'm a bit suspicious that they don't extend their perplexity numbers to the 13B model, or provide the hyper parameters, but they reference it in text and in their scaling table.

Code will be released in a week https://github.com/microsoft/unilm/tree/master/retnet