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
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 changed | Where | Measured | Effect |
|---|---|---|---|
| Use Ouroboros delayed-gradient model parallelism with Adam instead of sequential Adam | Transformer · enwiki8 test | bpc: 1.11 -> 1.12 | Hurt |
| Use Ouroboros delayed-gradient model parallelism with Adam instead of sequential Adam | Transformer · text8 test | bpc: 1.18 -> 1.18 | No clear effect |
| Use Ouroboros delayed-gradient model parallelism with Adam instead of sequential Adam | Transformer · WikiText-103 test | PPL: 28.32 -> 28.29 | Improved |
| Use Ouroboros delayed-gradient model parallelism with Adam instead of sequential Adam | Transformer-XL · enwiki8 test | bpc: 1.06 -> 1.05 | Improved |
| Use Ouroboros delayed-gradient model parallelism with Adam instead of sequential Adam | Transformer-XL · text8 test | bpc: 1.15 -> 1.13 | Improved |
| Use Ouroboros delayed-gradient model parallelism with Adam instead of sequential Adam | Transformer-XL · WikiText-103 test | PPL: 24.00 -> 24.10 | Hurt |
| Use Ouroboros model parallelism rather than sequential backpropagation | Transformer-based language models · enwiki8 and text8; WikiText-103 | training time | Improved |
| Remove early-stage warm-up (ablation) | Ouroboros + Adam · enwiki8 | training loss convergence | Hurt |
| Use warm-up steps 50, 500, or 5000 instead of no warm-up (ablation) | Ouroboros + Adam · enwiki8 | training loss convergence | Improved |
| Remove repeatable dropout (recompute activations with fresh dropout randomness) (ablation) | Ouroboros + Adam · enwiki8 | training loss convergence | Hurt |
Closest papers by meaning
- Rethinking Memory and Communication Costs for Efficient Data Parallel Training of Large Language Models (NeurIPS 2024 · similarity 0.88)
- Test-Time Training on Nearest Neighbors for Large Language Models (ICLR 2024 · similarity 0.87)
- Trainable Transformer in Transformer (ICML 2024 · similarity 0.87)
- Bigram Subnetworks: Mapping to Next Tokens in Transformer Language Models (NeurIPS 2025 · similarity 0.86)
- Memory-Augmenting Decoder-Only Language Models through Encoders (Student Abstract) (AAAI 2024 · similarity 0.86)
- Building on Efficient Foundations: Effective Training of LLMs with Structured Feedforward Layers (NeurIPS 2024 · similarity 0.86)
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.
