LEMON: Reviving Stronger and Smaller LMs from Larger LMs with Linear Parameter Fusion (2024.acl-long)
Copied to clipboard
| Challenge: | Existing methods to train a stronger and smaller model with the help of large models are limited by the model size and performance. |
| Approach: | They propose to learn competent initial points for smaller models by fusing parameters from larger models and introduce controllable receptive fields to model prior parameter characteristics. |
| Outcome: | The proposed method outperforms baselines in terms of effectiveness and efficiency. |
Similar Papers
Cool-Fusion: Fuse Large Language Models without Training (2025.acl-long)
Copied to clipboard
| Challenge: | Cool-Fusion is a simple yet effective approach to combine two or more heterogeneous large language models . |
| Approach: | They propose a method that fuses the knowledge of two or more heterogeneous large language models to leverage complementary strengths. |
| Outcome: | The proposed method increases accuracy from three strong source LLMs on GSM8K by 17.4%. |
LM-Cocktail: Resilient Tuning of Language Models via Model Merging (2024.findings-acl)
Copied to clipboard
| Challenge: | Pre-trained language models are continually fine-tuned to better support downstream applications. however, this operation may result in significant performance degeneration on general perspectives. |
| Approach: | They propose a method which enables pre-trained language models to stay resilient in general perspectives. |
| Outcome: | The proposed model achieves strong empirical performance in the whole scope of general tasks while preserving a superior capacity in its targeted domain. |
Small Language Models Improve Giants by Rewriting Their Outputs (2024.eacl-long)
Copied to clipboard
| Challenge: | despite impressive performance of large language models, they lag behind specialized models in various tasks. |
| Approach: | They propose a training model that can be integrated with different LLMs at inference to improve their performance without task-specific training. |
| Outcome: | The proposed model outperforms standard models on four natural language generation tasks. |
A Data-Efficient Path to Multilingual LLMs: Language Expansion via Post-training PARAMš„ Integration into Upcycled MoE (2026.acl-long)
Copied to clipboard
Hao Zhou, Tianhao Li, Zhijun Wang, Shuaijie She, Linjuan Wu, Hao-Ran Wei, Baosong Yang, Jiajun Chen, Shujian Huang
| Challenge: | Large Language Models (LLMs) are expensive and require extensive Continued Pre-Training and data-intensive alignment to expand. |
| Approach: | They propose a method which upcycles a dense model into a Mixture-of-Experts architecture, allocating different experts to different languages. |
| Outcome: | Experiments show that the proposed model upcycles a dense model into a Mixture-of-Experts(MoE) architecture, allocating different experts to different languages. |
Small Pre-trained Language Models Can be Fine-tuned as Large Models via Over-Parameterization (2023.acl-long)
Copied to clipboard
| Challenge: | Large pre-trained language models (PLMs) have shown remarkable performance in various natural language processing tasks, outperforming small PLMs by a large margin. |
| Approach: | They propose to scale up parameters of pre-trained language models only during fine-tuning to benefit from over-parameterization. |
| Outcome: | The proposed approach can significantly boost the fine-tuning performance of small PLMs and even help small PDMs outperform 3 parameterized larger ones. |
CoLA: Compute-Efficient Pre-Training of LLMs via Low-Rank Activation (2025.emnlp-main)
Copied to clipboard
Ziyue Liu, Ruijie Zhang, Zhengyang Wang, Mingsong Yan, Zi Yang, Paul D. Hovland, Bogdan Nicolae, Franck Cappello, Sui Tang, Zheng Zhang
| Challenge: | Large foundation models have become huge, but they consume computational resources in pretraining. |
| Approach: | They propose to replace full-size layers with compute-efficient auto-encoders that enforce low-rank activations throughout training. |
| Outcome: | The proposed method reduces the computing cost by 2pmbtimes and improves training throughput by 1.86pmtime. |
Mini-Model Adaptation: Efficiently Extending Pretrained Models to New Languages via Aligned Shallow Training (2023.findings-acl)
Copied to clipboard
| Challenge: | Existing approaches to pretrain Masked Language Models (MLMs) are expensive and require a full forward and backward pass over the entire model. |
| Approach: | They propose to learn a shallow mini-model from a fraction of a large model's parameters and plug it into a larger model for rapid cross-lingual transfer. |
| Outcome: | Experiments on XNLI, MLQA and PAWS-X show that mini-model adaptation matches the standard approach using up to 2.3x less compute on average. |
Plug-in and Fine-tuning: Bridging the Gap between Small Language Models and Large Language Models (2025.acl-long)
Copied to clipboard
| Challenge: | Large language models (LLMs) are renowned for their extensive linguistic knowledge and strong generalization capabilities, but their high computational demands make them unsuitable for resource-constrained environments. |
| Approach: | They propose a framework that integrates a single frozen layer from an LLM into a SLM and fine-tunes the combined model for specific tasks. |
| Outcome: | The proposed framework improves performance across a range of natural language processing tasks, including both natural language understanding and generation. |
Decoupling Generalization and Adaptation in Meta-Learning for Large Language Models (2026.acl-short)
Copied to clipboard
| Challenge: | Adapting large language models to specific downstream tasks requires multi-step fine-tuning with substantial training data, incurring significant computational overhead. |
| Approach: | They propose a framework that separates learning generalizable initializations and adaptation through dedicated parameter spaces. |
| Outcome: | The proposed framework outperforms existing meta-learning and standard multi-task baselines on common-sense reasoning, mathematics, logic, medical and coding benchmarks. |
LESA: Learnable LLM Layer Scaling-Up (2025.acl-long)
Copied to clipboard
| Challenge: | Existing methods for depth scaling-up rely on empirical heuristic rules for layer duplication, resulting in poor initialization and slower convergence during continual pre-training. |
| Approach: | They propose a method for learning latent parameters between layers by concatenating parameters from each layer and applying Singular Value Decomposition. |
| Outcome: | Experiments show that LESA outperforms baseline models with less than half the cost of existing methods. |