Research map / Architectures and efficiency / Spiking networks
Plug-and-Play Spiking Operators: Breaking the Nonlinearity Bottleneck in Spiking Transformers
Xinzhe Yuan, Xiang Peng, Bin Gu, Huan Xiong
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 |
|---|---|---|---|
| Replace Softmax/SiLU/RMSNorm with NLSpike operators in ANN LLMs without retraining | LLaMA-3-8B · 5 zero-shot tasks avg | Avg. accuracy: 0.730 -> -0.003 | No clear effect |
| Replace nonlinear operators with NLSpike in Qwen3-8B | Qwen3-8B · ArcC and ArcE | accuracy: +0.039 ArcC, +0.033 ArcE | Improved |
| Apply NLSpike after SpikeLLM conversion | SpikeLLM W2A16 · 5 zero-shot tasks avg | Avg. accuracy: -0.000 (7B), -0.001 (13B) | No clear effect |
| Apply NLSpike to SpikeZIP-converted BERT | BERT SpikeZIP T=64 · MR/SST-2/Subj/SST-5 | Avg. accuracy: 0.807 -> +0.003 | No clear effect |
| Increase softmax/SiLU clipping range H from 3 to 10 (SiLU) (ablation) | NLS-SiLU · SiLU approximation error | max error | Hurt |
| Increase clipping range H (Softmax, d=64) (ablation) | NLS-Softmax · Softmax approximation error | mean and max error | Improved |
| NLS-RMS versus blockwise RMS approximations | RMSNorm · Operator-level error, 8-bit | mean error | Improved |
| Larger time steps T (1 to 4) for NLS vs SNN (ablation) | NLS on LLaMA · 5 tasks avg | Avg. accuracy | No clear effect |
| NLS at T=1 W6A6 on Llama 2 7B | NLS · WinoGrande | accuracy: 70.09 -> 67.48 | Hurt |
Closest papers by meaning
- Spike-driven Transformer V2: Meta Spiking Neural Network Architecture Inspiring the Design of Next-generation Neuromorphic Chips (ICLR 2024 · similarity 0.90)
- STEP: A Unified Spiking Transformer Evaluation Platform for Fair and Reproducible Benchmarking (NeurIPS 2025 · similarity 0.89)
- Towards Efficient Spiking Transformer: a Token Sparsification Framework for Training and Inference Acceleration (ICML 2024 · similarity 0.89)
- Spiking Transformer: Introducing Accurate Addition-Only Spiking Self-Attention for Transformer (CVPR 2025 · similarity 0.89)
- Sorbet: A Neuromorphic Hardware-Compatible Transformer-Based Spiking Language Model (ICML 2025 · similarity 0.89)
- TTFSFormer: A TTFS-based Lossless Conversion of Spiking Transformer (ICML 2025 · similarity 0.89)
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.
