Research map / Architectures and efficiency / Spiking networks
SpikingLM: Towards Fully Spiking Language Model
Yu Liang, Zijian Zhou, Wenjie Wei, Shuai Wang, Honglin Cao, Ammar Belatreche, Yu Yang, Malu Zhang, Yang Yang, Haizhou Li
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 |
|---|---|---|---|
| add Spike2Max attention instead of spiking self-attention | SpikingLM backbone · Language modeling | perplexity: 24.58 -> 19.78 | Improved |
| replace SSA with Spike2Max | SpikingLM backbone · Language modeling | energy (mJ): 1.580 -> 1.457 | Improved |
| add Distribution-aware Scaling alone (ablation) | Base spiking BERT · Pre-training | perplexity: 24.47 | Improved |
| add Spike2Max alone without scaling (ablation) | Base spiking BERT · Pre-training | perplexity: 256.39 | Hurt |
| add Spike2Max on top of Distribution-aware Scaling (ablation) | SpikingLM · Pre-training | perplexity: 24.47 -> 19.78 | Improved |
| use Distribution-aware Scaling instead of RMSNorm (ablation) | SpikingLM · Pre-training | perplexity: 22.18 -> 19.78 | Improved |
| use Distribution-aware Scaling instead of RMSNorm (ablation) | SpikingLM · Pre-training | energy (mJ): 1.643 -> 1.545 | Improved |
| use learnable Distribution-aware Scaling instead of fixed scale 2 (ablation) | SpikingLM · Pre-training | perplexity: 45.93 -> 19.78 | Improved |
| use Spike2Max instead of standard Softmax (ablation) | SpikingLM · Pre-training | perplexity: 18.29 -> 19.78 | Hurt |
| use Spike2Max instead of standard Softmax (ablation) | SpikingLM · Pre-training | energy (mJ): 1.761 -> 1.457 | Improved |
| add learnable temporal gamma to Max-Sub (Spike2Max vs Max-Sub) (ablation) | SpikingLM · Pre-training | perplexity: 22.75 -> 19.78 | Improved |
| use Spike2Max/DAS SpikingLM instead of SpikeLM | GLUE · GLUE | average score: 76.5 -> 77.1 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Spikingformer — Spikingformer: A Key Foundation Model for Spiking Neural Networks (AAAI 2026) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- SpikeLLM: Scaling up Spiking Neural Network to Large Language Models via Saliency-based Spiking (ICLR 2025 · similarity 0.92)
- SpikeLM: Towards General Spike-Driven Language Modeling via Elastic Bi-Spiking Mechanisms (ICML 2024 · similarity 0.92)
- SpikingSSMs: Learning Long Sequences with Sparse and Parallel Spiking State Space Models (AAAI 2025 · similarity 0.90)
- Sorbet: A Neuromorphic Hardware-Compatible Transformer-Based Spiking Language Model (ICML 2025 · similarity 0.90)
- SpikingBERT: Distilling BERT to Train Spiking Language Models Using Implicit Differentiation (AAAI 2024 · similarity 0.89)
- QP-SNN: Quantized and Pruned Spiking Neural Networks (ICLR 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.
