Research map / Federated learning and privacy / Personalized and heterogeneous FL
CoLM: Collaborative Large Models via a Client-Server Paradigm
Siqi Huang, Sida Huang, Hongyuan Zhang
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 |
|---|---|---|---|
| apply CoLM collaboration | Janus-Pro-7B · VLM benchmarks average | Avg. Score: 47.55 -> 58.06 | Improved |
| apply CoLM collaboration | LLaVA-1.5-7B · VLM benchmarks average | Avg. Score: 38.62 -> 56.13 | Improved |
| apply CoLM collaboration | GPT-4o · VLM benchmarks average | Avg. Score: 63.51 -> 64.53 | Improved |
| apply CoLM collaboration | Qwen2.5-VL-7B · MME-P | score: 1693.53 -> 1656.04 | Hurt |
| apply CoLM collaboration | Qwen2.5-Math-7B-Instruct · LLM benchmarks average | Avg. Score: 15.02 -> 26.71 | Improved |
| apply CoLM collaboration | Deepseek-Math-7B-Instruct · LLM benchmarks average | Avg. Score: 15.53 -> 56.28 | Improved |
| apply CoLM collaboration | Qwen2.5-Coder-1.5B · MT-Bench Avg. | score: 2.30 -> 1.05 | Hurt |
| apply CoLM collaboration | Qwen2.5-Math-1.5B · Small LLMs aggregate | Agg. Score: 11.09 -> 11.58 | No clear effect |
| use server output instead of MoA aggregation | GPT-4o · LLM benchmarks average | Avg. Score: 82.91 -> 83.93 | Improved |
| collaborate with only Qwen2.5-VL-7B client (ablation) | Janus-Pro-7B · MMMU | accuracy: 0.373 -> 0.407 | Improved |
| collaborate with only LLaVA-1.5-7B client (ablation) | Janus-Pro-7B · MMMU | accuracy: 0.373 -> 0.260 | Hurt |
| collaborate with only GPT-4o client (ablation) | Janus-Pro-7B · MMMU | accuracy: 0.373 -> 0.533 | Improved |
Closest papers by meaning
- ACC-Collab: An Actor-Critic Approach to Multi-Agent LLM Collaboration (ICLR 2025 · similarity 0.88)
- Collab: Controlled Decoding using Mixture of Agents for LLM Alignment (ICLR 2025 · similarity 0.86)
- CollabLLM: From Passive Responders to Active Collaborators (ICML 2025 · similarity 0.86)
- CoRe: Collaborative Reasoning via Cross Teaching (ICML 2026 · similarity 0.86)
- Hogwild! Inference: Parallel LLM Generation via Concurrent Attention (NeurIPS 2025 · similarity 0.86)
- Insight-V: Exploring Long-Chain Visual Reasoning with Multimodal Large Language Models (CVPR 2025 · similarity 0.86)
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.
