Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Linearizing Large Language Models

About

Linear transformers have emerged as a subquadratic-time alternative to softmax attention and have garnered significant interest due to their fixed-size recurrent state that lowers inference cost. However, their original formulation suffers from poor scaling and underperforms compute-matched transformers. Recent linear models such as RWKV and Mamba have attempted to address these shortcomings by proposing novel time-mixing and gating architectures, but pre-training large language models requires significant data and compute investments. Thus, the search for subquadratic architectures is limited by the availability of compute and quality pre-training datasets. As a cost-effective alternative to pre-training linear transformers, we propose Scalable UPtraining for Recurrent Attention (SUPRA). We present a method to uptrain existing large pre-trained transformers into Recurrent Neural Networks (RNNs) with a modest compute budget. This allows us to leverage the strong pre-training data and performance of existing transformer LLMs, while requiring 5% of the training cost. We find that our linearization technique leads to competitive performance on standard benchmarks, but we identify persistent in-context learning and long-context modeling shortfalls for even the largest linear models. Our code and models can be found at https://github.com/TRI-ML/linear_open_lm.

Jean Mercat, Igor Vasiljevic, Sedrick Keh, Kushal Arora, Achal Dave, Adrien Gaidon, Thomas Kollar• 2024

Related benchmarks

TaskDatasetResultRank
Commonsense ReasoningWinoGrande
Accuracy70.3
1442
Question AnsweringARC Challenge
Accuracy45.7
906
Physical Commonsense ReasoningPIQA
Accuracy80.1
696
Question AnsweringARC Easy
Accuracy76
597
Multitask Language UnderstandingMMLU
Accuracy34.1
520
Sentence CompletionHellaSwag
Accuracy77.1
364
Multitask Language UnderstandingMMLU
Accuracy40.9
263
Question AnsweringARC-C--
116
Commonsense ReasoningWinoGrande
Accuracy70.3
103
Common Sense ReasoningPIQA
Accuracy80.4
100
Showing 10 of 22 rows

Other info

Follow for update