Research map / Large language models / Language, safety and interpretability
Learn-to-learn on Arbitrary Textual Conditioning: A Hypernetwork-Driven Meta-gated LLM
Luo Ji, Qi Qin, Ningyuan Xi, Teng Chen, Qingqing Gu, Hongyan 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 the meta-prompt p as hypernetwork input (ablation) | MeGan · AdaptSum | ROUGE-L: 0.18 -> 41.85 | Improved |
| add the meta-prompt p as hypernetwork input (ablation) | MeGan · Persona-Chat | ROUGE-L: 20.38 -> 23.15 | Improved |
| add positional embedding in the hypernetwork (ablation) | MeGan · Persona-Chat | ROUGE-L: 7.11 -> 23.15 | Improved |
| add positional embedding in the hypernetwork (ablation) | MeGan · AdaptSum | ROUGE-L: 16.82 -> 41.85 | Improved |
| add L2 regularization on beta (ablation) | MeGan · AdaptSum | ROUGE-L: 36.62 -> 41.85 | Improved |
| add L2 regularization on beta (ablation) | MeGan · Persona-Chat | ROUGE-L: 20.38 -> 23.15 | Improved |
| add L2 regularization on beta (ablation) | MeGan · Persona-Chat | BLEU-2: 10.15 -> 10.15 | No clear effect |
| replace Text-to-LoRA with MeGan meta-gating | Llama-3.1-8B-Instruct · Persona-Chat | BLEU-2: 13.70 -> 10.15 | Hurt |
| replace LoRA with MeGan meta-gating | Llama-3.1-8B-Instruct · SST | ROUGE-L: 13.31 -> 22.76 | Improved |
| replace Text-to-LoRA with MeGan meta-gating | Llama-3.1-8B-Instruct · CrossFit&UnifiedQA HR->LR (entire test set) | average performance: 64.7 -> 66.7 | Improved |
| replace Text-to-LoRA with MeGan meta-gating | Llama-3.1-8B-Instruct · SNI-trained zero-shot, average of 10 benchmarks | average score: 76.9 -> 77.5 | Improved |
| replace Text-to-LoRA with MeGan meta-gating | Llama-3.1-8B-Instruct · GSM8K (SNI-trained) | accuracy: 79.1 -> 74.1 | Hurt |
Closest papers by meaning
- Prompt-MII: Meta-Learning Instruction Induction for LLMs (ICLR 2026 · similarity 0.87)
- Improving Generalization of Meta-Learning With Inverted Regularization at Inner-Level (CVPR 2023 · similarity 0.87)
- Dynamic Sparse No Training: Training-Free Fine-tuning for Sparse LLMs (ICLR 2024 · similarity 0.87)
- Drag-and-Drop LLMs: Zero-Shot Prompt-to-Weights (NeurIPS 2025 · similarity 0.87)
- Mitigating Forgetting in LLM Fine-Tuning via Low-Perplexity Token Learning (NeurIPS 2025 · similarity 0.85)
- Learn To be Efficient: Build Structured Sparsity in Large Language Models (NeurIPS 2024 · 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.
