DG-Hard uses Donoho-Gavish hard thresholding on the fine-tuning weight delta to separate task-aligned signal from noise-like residual, recovering damaged capabilities while preserving target-task gains.
Balancing speciality and versatility: A coarse to fine framework for mitigating catastrophic forgetting in large language models
7 Pith papers cite this work. Polarity classification is still indexing.
7
Pith papers citing it
abstract
Aligned Large Language Models (LLMs) showcase remarkable versatility, capable of handling diverse real-world tasks. Meanwhile, aligned LLMs are also expected to exhibit speciality, excelling in specific applications. However, fine-tuning with extra data, a common practice to gain speciality, often leads to catastrophic forgetting (CF) of previously acquired versatility, hindering the model's performance across diverse tasks. In response to this challenge, we propose CoFiTune, a coarse to fine framework in an attempt to strike the balance between speciality and versatility. At the coarse-grained level, an empirical tree-search algorithm is utilized to pinpoint and update specific modules that are crucial for speciality, while keeping other parameters frozen; at the fine-grained level, a soft-masking mechanism regulates the update to the LLMs, mitigating the CF issue without harming speciality. In an overall evaluation of both speciality and versatility, CoFiTune consistently outperforms baseline methods across diverse tasks and model scales. Compared to the full-parameter SFT, CoFiTune leads to about 14% versatility improvement and marginal speciality loss on a 13B model. Lastly, based on further analysis, we provide a speculative insight into the information forwarding process in LLMs, which helps explain the effectiveness of the proposed method. The code is available at https://github.com/rattlesnakey/CoFiTune.
citation-role summary
background 1
citation-polarity summary
roles
background 1polarities
background 1representative citing papers
The paper introduces Neuron-Guided Fine-Tuning (NGFT), a framework that uses neuron activation patterns to unify data selection, parameter updates, and knowledge preservation during LLM fine-tuning.
Adding a KL penalty between fine-tuned and original model logits on input tokens during RAG fine-tuning reduces catastrophic forgetting while preserving downstream performance.
IDEAL tunes SFT data mixture proportions per domain with influence-function gradients, claiming about 7% average benchmark improvement over uniform mixing.
BPO balances knowledge breadth and depth in preference data by compressing prompts and dynamically augmenting the number of response pairs per prompt using gradient-based clustering, achieving stronger alignment with roughly 10% of the data.
FAPM prunes task vectors using a combined absolute and relative change magnitude criterion, reducing forgetting on general benchmarks to roughly a third of a percent while keeping downstream accuracy near the fine-tuned model on Llama3-8B and Qwen2-7B.
A survey on LLM-as-a-Judge that reviews reliability strategies, proposes evaluation methods, and introduces a novel benchmark for assessing such systems.
citing papers explorer
-
Spectral Unforgetting: Post-Hoc Recovery of Damaged Capabilities Without Retraining
DG-Hard uses Donoho-Gavish hard thresholding on the fine-tuning weight delta to separate task-aligned signal from noise-like residual, recovering damaged capabilities while preserving target-task gains.
-
Neuron-Guided Fine-Tuning: Unlocking Efficient Alignment Mechanisms for Large Language Models
The paper introduces Neuron-Guided Fine-Tuning (NGFT), a framework that uses neuron activation patterns to unify data selection, parameter updates, and knowledge preservation during LLM fine-tuning.
-
SelfAug: Mitigating Catastrophic Forgetting in Retrieval-Augmented Generation via Distribution Self-Alignment
Adding a KL penalty between fine-tuned and original model logits on input tokens during RAG fine-tuning reduces catastrophic forgetting while preserving downstream performance.
-
IDEAL: Data Equilibrium Adaptation for Multi-Capability Language Model Alignment
IDEAL tunes SFT data mixture proportions per domain with influence-function gradients, claiming about 7% average benchmark improvement over uniform mixing.
-
BPO: Towards Balanced Preference Optimization between Knowledge Breadth and Depth in Alignment
BPO balances knowledge breadth and depth in preference data by compressing prompts and dynamically augmenting the number of response pairs per prompt using gradient-based clustering, achieving stronger alignment with roughly 10% of the data.
-
Mitigating Catastrophic Forgetting in Large Language Models with Forgetting-aware Pruning
FAPM prunes task vectors using a combined absolute and relative change magnitude criterion, reducing forgetting on general benchmarks to roughly a third of a percent while keeping downstream accuracy near the fine-tuned model on Llama3-8B and Qwen2-7B.
-
A Survey on LLM-as-a-Judge
A survey on LLM-as-a-Judge that reviews reliability strategies, proposes evaluation methods, and introduces a novel benchmark for assessing such systems.