Research map / Architectures and efficiency / Pruning and efficient tuning
Zero-Shot Sparse Mixture of Low-Rank Experts Construction From Pre-Trained Foundation Models
Anke Tang, Li Shen, Yong Luo, Shuai Xie, Han Hu, Lefei Zhang, Bo Du, Dacheng Tao
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 SMILE (k_gate=16, k=128) instead of Task Arithmetic | CLIP-ViT-B/32 fusion · 8 image tasks (CLIP-ViT-B/32) | average accuracy: 68.0 -> 89.3 | Improved |
| use SMILE (k_gate=16, k=128) instead of AdaMerging | CLIP-ViT-L/14 fusion · 8 image tasks (CLIP-ViT-L/14) | average accuracy: 91.0 -> 93.6 | Improved |
| use SMILE (2.56x params) instead of Weight-Ensembling MoE (6.40x params) | CLIP-ViT-L/14 fusion · 8 image tasks (CLIP-ViT-L/14) | average accuracy: 93.8 -> 93.6 | Mixed |
| increase expert rank k from 32 to 128 (ablation) | SMILE · CLIP-ViT-B/32 | average accuracy: 87.7 -> 89.3 | Improved |
| increase expert rank k from 32 to 128 (ablation) | SMILE · CLIP-ViT-L/14 | average accuracy: 92.8 -> 93.6 | Improved |
| increase k and k_gate (k=16,k_gate=4 to k=32,k_gate=8) (ablation) | SMILE · GLUE full fine-tuned Flan-T5-Base | average score: 85.1 -> 85.5 | Improved |
| use SMILE instead of Ties-Merging | Flan-T5-Base LoRA fusion · GLUE 8 tasks (LoRA) | average score: 77.5 -> 84.0 | Improved |
| increase expert rank k from 8 to 512 (ablation) | SMILE Mistral-7B · GSM8K | accuracy: 46.55 -> 67.85 | Improved |
| increase Top-K in routing (ablation) | SMILE · CLIP-ViT-B/32 (k_gate=16, k=32) | average accuracy: average accuracy decreases slightly as K increases | Hurt |
| increase k_gate beyond moderate values (ablation) | SMILE · CLIP-ViT models | average accuracy | Hurt |
| merge into SMILE (M0;123) versus individual math expert M1 | SMILE Mistral-7B · GSM8K | accuracy: 71.49 -> 67.85 | Mixed |
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.
- Ties-Merging — TIES-Merging: Resolving Interference When Merging Models (NeurIPS 2023) · better on 4 of 4 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
- Unleashing the Power of Meta-tuning for Few-shot Generalization Through Sparse Interpolated Experts (ICML 2024 · similarity 0.87)
- Mitigating Parameter Interference in Model Merging via Sharpness-Aware Fine-Tuning (ICLR 2025 · similarity 0.86)
- Sparse Model Soups: A Recipe for Improved Pruning via Model Averaging (ICLR 2024 · similarity 0.86)
- DeRS: Towards Extremely Efficient Upcycled Mixture-of-Experts Models (CVPR 2025 · similarity 0.86)
- A Provably Effective Method for Pruning Experts in Fine-tuned Sparse Mixture-of-Experts (ICML 2024 · similarity 0.86)
- DARE the Extreme: Revisiting Delta-Parameter Pruning For Fine-Tuned Models (ICLR 2025 · similarity 0.85)
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.
