AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Ouroboros: On Accelerating Training of Transformer-Based Language Models

Qian Yang, Zhouyuan Huo, Wenlin Wang, Lawrence Carin

NeurIPS 2019 · Architectures and efficiency › Attention and transformers · approach: Efficient language transformers · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Use Ouroboros delayed-gradient model parallelism with Adam instead of sequential AdamTransformer · enwiki8 testbpc: 1.11 -> 1.12Hurt
Use Ouroboros delayed-gradient model parallelism with Adam instead of sequential AdamTransformer · text8 testbpc: 1.18 -> 1.18No clear effect
Use Ouroboros delayed-gradient model parallelism with Adam instead of sequential AdamTransformer · WikiText-103 testPPL: 28.32 -> 28.29Improved
Use Ouroboros delayed-gradient model parallelism with Adam instead of sequential AdamTransformer-XL · enwiki8 testbpc: 1.06 -> 1.05Improved
Use Ouroboros delayed-gradient model parallelism with Adam instead of sequential AdamTransformer-XL · text8 testbpc: 1.15 -> 1.13Improved
Use Ouroboros delayed-gradient model parallelism with Adam instead of sequential AdamTransformer-XL · WikiText-103 testPPL: 24.00 -> 24.10Hurt
Use Ouroboros model parallelism rather than sequential backpropagationTransformer-based language models · enwiki8 and text8; WikiText-103training timeImproved
Remove early-stage warm-up (ablation)Ouroboros + Adam · enwiki8training loss convergenceHurt
Use warm-up steps 50, 500, or 5000 instead of no warm-up (ablation)Ouroboros + Adam · enwiki8training loss convergenceImproved
Remove repeatable dropout (recompute activations with fresh dropout randomness) (ablation)Ouroboros + Adam · enwiki8training loss convergenceHurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy