Research map / Architectures and efficiency / Attention and transformers
ParaDySe: A Parallel Strategy Switching Framework for Dynamic Sequences in Transformer-based Large Language Models
Zhixin Ou, Peng Liang, Linbo Qiao, Jianchen Han, Baihui Liu
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 ParaDySe adaptive strategy switching instead of static baselines on GitHubCode | BERT training · GitHubCode max sequences | training time: up to 89.29% less time | Improved |
| use ParaDySe instead of baselines on GitHubCode | GPT (largest) · GitHubCode | max trainable sequence length: up to +144.16% | Improved |
| use ParaDySe instead of baselines on GRCh38 | BERT / GPT · GRCh38 | training time / sequence length: 58.01% faster (BERT); 181.22% longer (GPT) | Improved |
| use ParaDySe instead of baselines for LLaMA on one case | LLaMA · long-sequence datasets | max sequence length | Mixed |
| add MegatronTS to the strategy set (ablation) | ParaDySe · GPT on GRCh38 | time saving: 300032 -> 329728 tokens; 11.80% time saving | Improved |
| add METP to the strategy set (ablation) | ParaDySe · GPT on GRCh38 | max sequence length: 112128 -> 329728 tokens | Improved |
| add MegatronCZ to the strategy set (ablation) | ParaDySe · GPT on GRCh38 | time saving: -0.13% saving; seq len 300032 -> 329728 | No clear effect |
| add random forest to the polynomial cost model (hybrid) (ablation) | ParaDySe cost model · GPT on GRCh38 | max sequence length: 270336 -> 329728 tokens; 1.29% saving | Improved |
| add smoothing of strategy switching (ablation) | ParaDySe · GPT on GRCh38 | time saving: 0.04% saving; 300032 -> 329728 tokens | Improved |
Closest papers by meaning
- MISA: Memory-Efficient LLMs Optimization with Module-wise Importance Sampling (NeurIPS 2025 · similarity 0.87)
- Relaxed Recursive Transformers: Effective Parameter Sharing with Layer-wise LoRA (ICLR 2025 · similarity 0.87)
- Rethinking Memory and Communication Costs for Efficient Data Parallel Training of Large Language Models (NeurIPS 2024 · similarity 0.87)
- Benchmarking Optimizers for Large Language Model Pretraining (NeurIPS 2026 · similarity 0.87)
- Enhancing Large Language Model Performance with Gradient-Based Parameter Selection (AAAI 2025 · similarity 0.87)
- Joint Fine-tuning and Conversion of Pretrained Speech and Language Models towards Linear Complexity (ICLR 2025 · 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.
