paper-with-me

홈 › Papers

Benchmarking down-scaled (not so large) pre-trained language models

2021-05-11 · M. Aßenmacher, P. Schulze, C. Heumann

Large Transformer-based language models are pre-trained on corpora of varying sizes, for a different number of steps and with different batch sizes. At the same time, more fundamental components, such as the pre-training objective or architectural hyperparameters, are modified. In total, it is therefore difficult to ascribe changes in performance to specific factors. Since searching the hyperparameter space over the full systems is too costly, we pre-train down-scaled versions of several popular Transformer-based architectures on a common pre-training corpus and benchmark them on a subset of the GLUE tasks (Wang et al., 2018). Specifically, we systematically compare three pre-training objectives for different shape parameters and model sizes, while also varying the number of pre-training steps and the batch size. In our experiments MLM + NSP (BERT-style) consistently outperforms MLM (RoBERTa-style) as well as the standard LM objective. Furthermore, we find that additional compute should be mainly allocated to an increased model size, while training for more steps is inefficient. Based on these observations, as a final step we attempt to scale up several systems using compound scaling (Tan and Le, 2019) adapted to Transformer-based language models.

📄 PDF Abstract BibTeX arXiv:2105.04876

Code (1)

PMSchulze/NLP-benchmarking 공식 구현 pytorch

Tasks

Benchmarking

Similar Papers 제목 키워드 기반

Benchmarking down-scaled (not so large) pre-trained language models

2021-09-01 · KONVENS (WS) 2021 9 · Matthias Aßenmacher, Patrick Schulze, Christian Heumann
Benchmarking

Weight subcloning: direct initialization of transformers using larger pretrained ones

2023-12-14 · Mohammad Samragh, Mehrdad Farajtabar, Sachin Mehta, Raviteja Vemulapalli 외

Training large transformer models from scratch for a target task requires lots of data and is computationally demanding. The usual practice of transfer learning overcomes this challenge by initializing the model with wei…

image-classificationImage ClassificationTransfer Learning

ScaleDet: A Scalable Multi-Dataset Object Detector

2023-06-08 · CVPR 2023 1 · Yanbei Chen, Manchen Wang, Abhay Mittal, Zhenlin Xu 외

Multi-dataset training provides a viable solution for exploiting heterogeneous large-scale datasets without extra annotation cost. In this work, we propose a scalable multi-dataset detector (ScaleDet) that can scale up i…

Objectobject-detectionObject Detection

Downscaled Representation Matters: Improving Image Rescaling with Collaborative Downscaled Images

2022-11-19 · ICCV 2023 1 · Bingna Xu, Yong Guo, Luoqian Jiang, Mianjie Yu 외

Deep networks have achieved great success in image rescaling (IR) task that seeks to learn the optimal downscaled representations, i.e., low-resolution (LR) images, to reconstruct the original high-resolution (HR) images…

Image ReconstructionImage RescalingSuper-Resolution

Blind Super-Resolution Kernel Estimation using an Internal-GAN

2019-09-14 · NeurIPS 2019 12 · Sefi Bell-Kligler, Assaf Shocher, Michal Irani

Super resolution (SR) methods typically assume that the low-resolution (LR) image was downscaled from the unknown high-resolution (HR) image by a fixed 'ideal' downscaling kernel (e.g. Bicubic downscaling). However, this…

Blind Super-ResolutionSuper-Resolution