Research map / Large language models / Language, safety and interpretability
LAS: Loss-less ANN-SNN Conversion for Fully Spike-Driven Large Language Models
Long Chen, Xiaotian Song, Yanan Sun
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 OAT neuron with a single multi-threshold neuron (ablation) | LAS · RealWorldQA, Qwen2-VL-7B | accuracy: -3.36 drop without OAT | Improved |
| add spike-equivalent self-attention (ablation) | LAS · RealWorldQA, Qwen2-VL-7B | accuracy: 15.77 -> 66.79 | Improved |
| use 16 time steps instead of 11 (ablation) | LAS · GLUE QNLI, BERT-base | accuracy: 86.14 -> 90.92 | Improved |
| use 13 instead of 16 time steps (ablation) | LAS · GLUE QNLI, BERT-base | accuracy: 90.92 -> 90.91 | No clear effect |
| increase MT neuron threshold levels H from 1 to 10 (ablation) | LAS · BERT STS-B | energy ratio: 1.03 -> 0.41 | Improved |
| increase H (ablation) | LAS · BERT STS-B | accuracy: 88.73 to 88.80 | No clear effect |
| convert OPT-66B to SNN with LAS | OPT-66B · WSC (zero-shot) | accuracy: 82.78 -> 85.71 | Improved |
| convert OPT-66B to SNN with LAS | OPT-66B · Winogrande (zero-shot) | accuracy: 69.98 -> 68.27 | Hurt |
| convert LLaVA1.5-7B with LAS | LLaVA1.5-7B · HallusionBench | score: 81.36 -> 67.93 | Hurt |
| convert Qwen2-VL-7B with LAS | Qwen2-VL-7B · BLINK, MMStar | accuracy: 51.55 -> 52.18 (BLINK) | Improved |
| use LAS instead of FAS conversion on OPT-7B | FAS · PIQA (zero-shot) | accuracy: 73.23 -> 76.22 | Improved |
Closest papers by meaning
- SpikeLLM: Scaling up Spiking Neural Network to Large Language Models via Saliency-based Spiking (ICLR 2025 · similarity 0.88)
- Inference-Scale Complexity in ANN-SNN Conversion for High-Performance and Low-Power Applications (CVPR 2025 · similarity 0.88)
- LM-HT SNN: Enhancing the Performance of SNN to ANN Counterpart through Learnable Multi-hierarchical Threshold Model (NeurIPS 2024 · similarity 0.88)
- TTFSFormer: A TTFS-based Lossless Conversion of Spiking Transformer (ICML 2025 · similarity 0.88)
- SpikedAttention: Training-Free and Fully Spike-Driven Transformer-to-SNN Conversion with Winner-Oriented Spike Shift for Softmax Operation (NeurIPS 2024 · similarity 0.87)
- Faster and Stronger: When ANN-SNN Conversion Meets Parallel Spiking Calculation (ICML 2025 · 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.
