Research map / Architectures and efficiency / Mixture of experts
cMoLLM at Scale: Horizontal Scaling Laws for Convolutionally-Gated Mixture-of-LLMs
Xin Yang, Yemin Wang, Mingda Liu, Letian Li, Shuaishuai Cao, ZhengXiao He, Ryan Dong
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 dense FFN with cMoLLM streams (multi-head, 8 streams) | GPT-2 85M · FineWeb validation | perplexity: 25.40 -> 18.82 | Improved |
| increase the number of streams from 1 to 8 (multi-head gating) (ablation) | cMoLLM multi-head · FineWeb validation, 85M | perplexity: 20.93 -> 18.82 | Improved |
| increase simple-gating streams from 4 to 8 (ablation) | cMoLLM simple · FineWeb validation, 85M | perplexity: 19.53 -> 20.24 | Hurt |
| use context-aware gating instead of simple gating at 8 streams (ablation) | cMoLLM · FineWeb GLUE, 85M | GLUE accuracy (%): 44.68 -> 48.98 | Improved |
| multi-head cMoLLM vs dense baseline at 350M | GPT-2 350M · FineWeb validation | perplexity: 16.80 -> 14.03 | Improved |
| multi-head cMoLLM vs dense baseline at 760M | GPT-2 760M · FineWeb validation | perplexity: 12.79 -> 9.41 | Improved |
| multi-head cMoLLM vs dense baseline at 350M | GPT-2 350M · GLUE | GLUE accuracy (%): 50.68 -> 53.82 | Improved |
| multi-head cMoLLM vs dense baseline at 760M | GPT-2 760M · SQuAD v2 | SQuAD v2 (%): 62.91 -> 64.56 | Improved |
| single stream (n=1) gated convolution vs dense FFN (ablation) | GPT-2 85M · FineWeb validation | perplexity: 25.40 -> 20.94 | Improved |
| increase streams from 1 to 8 with simple gating on SQuAD v2 (ablation) | cMoLLM simple · SQuAD v2, 85M | SQuAD v2 (%): 51.13 -> 55.95 | Improved |
Closest papers by meaning
- $\textit{Read-ME}$: Refactorizing LLMs as Router-Decoupled Mixture of Experts with System Co-Design (NeurIPS 2024 · similarity 0.89)
- $\texttt{Model-GLUE}$: Democratized LLM Scaling for A Large Model Zoo in the Wild (NeurIPS 2024 · similarity 0.88)
- Polar Sparsity: High Throughput Batched LLM Inferencing with Scalable Contextual Sparsity (NeurIPS 2025 · similarity 0.88)
- No Need to Talk: Asynchronous Mixture of Language Models (ICLR 2025 · similarity 0.88)
- Mini-batch Coresets for Memory-efficient Language Model Training on Data Mixtures (ICLR 2025 · similarity 0.88)
- Scaling Laws for Upcycling Mixture-of-Experts Language Models (ICML 2025 · similarity 0.88)
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.
