Research map / Architectures and efficiency / Attention and transformers
Sequential Parallel Duality in Prefix Scannable Models
Morris Yau, Sharut Gupta, Valerie Engelmayer, Kazuki Irie, Stefanie Jegelka, Jacob Andreas
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 |
|---|---|---|---|
| increase Transformer-PSM chunk size from 32 to 256 (ablation) | Transformer-PSM · WikiText-103 test | perplexity: 24.12 -> 22.45 | Improved |
| use Transformer-PSM (chunk 256) instead of Mamba 130M | language modeling · WikiText-103 test | perplexity: 24.7 -> 22.45 | Improved |
| use Transformer-PSM instead of full-context GPT-2 on perplexity | language modeling · WikiText-103 test | perplexity: 22.28 -> 22.45 | No clear effect |
| reduce Transformer-PSM chunk size from 64 to 32 on MQAR (ablation) | Transformer-PSM · MQAR length 512 | error rate | Hurt |
| use Transformer-PSM instead of Transformer or Mamba for S5 state tracking | sequence models trained on lengths up to 18 · S5 state tracking, length up to 180 | error rate | Improved |
| use Transformer-PSM instead of Mamba on MQAR with uniform queries | MQAR (8192 vocab, 8 key-value pairs) · MQAR | error rate | Improved |
| use Transformer-PSM instead of GPT-2 with KV cache for inference | inference latency · WikiText-2 per-token latency | inference time per token (s): 0.04 -> 0.008 | Improved |
Closest papers by meaning
- Parallelizing non-linear sequential models over the sequence length (ICLR 2024 · similarity 0.87)
- B'MOJO: Hybrid State Space Realizations of Foundation Models with Eidetic and Fading Memory (NeurIPS 2024 · similarity 0.87)
- RAT: Bridging RNN Efficiency and Attention Accuracy via Chunk-based Sequence Modeling (NeurIPS 2025 · similarity 0.87)
- Linear Attention for Efficient Bidirectional Sequence Modeling (NeurIPS 2025 · similarity 0.87)
- FFN Fusion: Rethinking Sequential Computation in Large Language Models (NeurIPS 2025 · similarity 0.87)
- Understanding the Differences in Foundation Models: Attention, State Space Models, and Recurrent Neural Networks (NeurIPS 2024 · similarity 0.87)
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.
