Safety alignment is an important procedure before the official deployment of a Large Language Model (LLM). While safety alignment has been extensively studied for LLM, there is still a large research gap for Large Reasoning Models (LRMs) that equip with improved reasoning capability. We in this paper systematically examine a simplified pipeline for producing safety aligned LRMs. With our evaluation of various LRMs, we deliver two main findings: i) Safety alignment can be done upon the LRM to restore its safety capability. ii) Safety alignment leads to a degradation of the reasoning capability of LRMs. The two findings show that there exists a trade-off between reasoning and safety capability with the sequential LRM production pipeline. The discovered trade-off, which we name Safety Tax, should shed light on future endeavors of safety research on LRMs. As a by-product, we curate a dataset called DirectRefusal, which might serve as an alternative dataset for safety alignment. Our source code is available at https://github.com/git-disl/Safety-Tax.

The advancement of Large Language Models (LLMs) and their finetuning strategies has triggered the renewed interests in multi-agent reinforcement learning. In this paper, we introduce a focal diversity-optimized multi-agent reinforcement learning approach, coined as MARL-Focal, with three unique characteristics. First, we develop an agent-fusion framework for encouraging multiple LLM based agents to collaborate in producing the final inference output for each LLM query. Second, we develop a focal-diversity optimized agent selection algorithm that can choose a small subset of the available agents based on how well they can complement one another to generate the query output. Finally, we design a conflict-resolution method to detect output inconsistency among multiple agents and produce our MARL-Focal output through reward-aware and policy-adaptive inference fusion. Extensive evaluations on five benchmarks show that MARL-Focal is cost-efficient and adversarial-robust. Our multi-agent fusion model achieves performance improvement of 5.51\% compared to the best individual LLM-agent and offers stronger robustness over the TruthfulQA benchmark. Code is available at https://github.com/sftekin/rl-focal

Recent research shows that Large Language Models (LLMs) are vulnerable to harmful fine-tuning attacks -- models lose their safety alignment ability after fine-tuning on a few harmful samples. For risk mitigation, a guardrail is typically used to filter out harmful samples before fine-tuning. By designing a new red-teaming method, we in this paper show that purely relying on the moderation guardrail for data filtration is not reliable. Our proposed attack method, dubbed Virus, easily bypasses the guardrail moderation by slightly modifying the harmful data. Experimental results show that the harmful data optimized by Virus is not detectable by the guardrail with up to 100\% leakage ratio, and can simultaneously achieve superior attack performance. Finally, the key message we want to convey through this paper is that: \textbf{it is reckless to consider guardrail moderation as a clutch at straws towards harmful fine-tuning attack}, as it cannot solve the inherent safety issue of the pre-trained LLMs. Our code is available at https://github.com/git-disl/Virus

Recent research demonstrates that the nascent fine-tuning-as-a-service business model exposes serious safety concerns -- fine-tuning over a few harmful data uploaded by the users can compromise the safety alignment of the model. The attack, known as harmful fine-tuning attack, has raised a broad research interest among the community. However, as the attack is still new, \textbf{we observe that there are general misunderstandings within the research community.} To clear up concern, this paper provide a comprehensive overview to three aspects of harmful fine-tuning: attacks setting, defense design and evaluation methodology. Specifically, we first present the threat model of the problem, and introduce the harmful fine-tuning attack and its variants. Then we systematically survey the existing literature on attacks/defenses/mechanical analysis of the problem. Finally, we introduce the evaluation methodology and outline future research directions that might contribute to the development of the field. Additionally, we present a list of questions of interest, which might be useful to refer to when reviewers in the peer review process question the realism of the experiment/attack/defense setting. A curated list of relevant papers is maintained and made accessible at: https://github.com/git-disl/awesome_LLM-harmful-fine-tuning-papers.

Alignment of pretrained LLMs using instruction-based datasets is critical for creating fine-tuned models that reflect human preference. A growing number of alignment-based fine-tuning algorithms and benchmarks emerged recently, fueling the efforts on effective alignments of pre-trained LLMs to ensure helpful, harmless, and honest answers from both open-source and closed-source LLMs. This paper tackles this problem by developing an alignment fusion approach, coined as $H^3$Fusion, with three unique characteristics. First, $H^3$Fusion ensembles multiple individually aligned LLMs to create a final fine-tuned alignment model with enhanced capabilities beyond those of individual models, delivering robust alignment through promoting helpful, harmless, honest fusion. Second, $H^3$Fusion leverages the mixture-of-experts (MoE) methodology in two steps. We first freeze the multi-head attention weights of each individual model while tuning the FFN layer during alignment fusion. Then we merge the aligned model weights with an expert router according to the type of input instruction and dynamically select a subset of experts that are best suited for producing the output response. Finally, we boost the performance of the resulting $H^3$3Fusion model by introducing gating loss and regularization terms. The former penalizes the selection errors of the expert-router, and the latter mediates the expert weights drifting during fine-tuning and dynamically adjusts the fusion behavior of the resulting model by canalizing the activations on the experts. Extensive evaluations on three benchmark datasets show that $H^3$3Fusion is more helpful, less harmful, and more honest from two aspects: it outperforms each individually aligned model by $11.37\%$, and it provides stronger robustness compared to the state-of-the-art LLM ensemble approaches by $13.77\%$. Code is available at github.com/sftekin/h3fusion.

Combining large language models during training or at inference time has shown substantial performance gain over component LLMs. This paper presents LLM-TOPLA, a diversity-optimized LLM ensemble method with three unique properties: (i) We introduce the focal diversity metric to capture the diversity-performance correlation among component LLMs of an ensemble. (ii) We develop a diversity-optimized ensemble pruning algorithm to select the top-k sub-ensembles from a pool of $N$ base LLMs. Our pruning method recommends top-performing LLM subensembles of size $S$, often much smaller than $N$. (iii) We generate new output for each prompt query by utilizing a learn-to-ensemble approach, which learns to detect and resolve the output inconsistency among all component LLMs of an ensemble. Extensive evaluation on four different benchmarks shows good performance gain over the best LLM ensemble methods: (i) In constrained solution set problems, LLM-TOPLA outperforms the best-performing ensemble (Mixtral) by 2.2\% in accuracy on MMLU and the best-performing LLM ensemble (MoreAgent) on GSM8k by 2.1\%. (ii) In generative tasks, LLM-TOPLA outperforms the top-2 performers (Llama70b/Mixtral) on SearchQA by $3.9\mathrm{x}$ in F1, and on XSum by more than $38$ in ROUGE-1. Our code and dataset, which contains outputs of 8 modern LLMs on 4 benchmarks is available at https://github.com/git-disl/llm-topla

Recent studies show that Large Language Models (LLMs) with safety alignment can be jail-broken by fine-tuning on a dataset mixed with harmful data. First time in the literature, we show that the jail-broken effect can be mitigated by separating states in the finetuning stage to optimize the alignment and user datasets. Unfortunately, our subsequent study shows that this simple Bi-State Optimization (BSO) solution experiences convergence instability when steps invested in its alignment state is too small, leading to downgraded alignment performance. By statistical analysis, we show that the \textit{excess drift} towards consensus could be a probable reason for the instability. To remedy this issue, we propose \textbf{L}azy(\textbf{i}) \textbf{s}afety \textbf{a}lignment (\textbf{Lisa}), which introduces a proximal term to constraint the drift of each state. Theoretically, the benefit of the proximal term is supported by the convergence analysis, wherein we show that a sufficient large proximal factor is necessary to guarantee Lisa's convergence. Empirically, our results on four downstream finetuning tasks show that Lisa with a proximal term can significantly increase alignment performance while maintaining the LLM's accuracy on the user tasks. Code is available at \url{https://github.com/git-disl/Lisa}.

This paper presents FusionShot, a focal diversity optimized few-shot ensemble learning approach for boosting the robustness and generalization performance of pre-trained few-shot models. The paper makes three original contributions. First, we explore the unique characteristics of few-shot learning to ensemble multiple few-shot (FS) models by creating three alternative fusion channels. Second, we introduce the concept of focal error diversity to learn the most efficient ensemble teaming strategy, rather than assuming that an ensemble of a larger number of base models will outperform those sub-ensembles of smaller size. We develop a focal-diversity ensemble pruning method to effectively prune out the candidate ensembles with low ensemble error diversity and recommend top-$K$ FS ensembles with the highest focal error diversity. Finally, we capture the complex non-linear patterns of ensemble few-shot predictions by designing the learn-to-combine algorithm, which can learn the diverse weight assignments for robust ensemble fusion over different member models. Extensive experiments on representative few-shot benchmarks show that the top-K ensembles recommended by FusionShot can outperform the representative SOTA few-shot models on novel tasks (different distributions and unknown at training), and can prevail over existing few-shot learners in both cross-domain settings and adversarial settings. For reproducibility purposes, FusionShot trained models, results, and code are made available at https://github.com/sftekin/fusionshot

The development of game agents holds a critical role in advancing towards Artificial General Intelligence (AGI). The progress of LLMs and their multimodal counterparts (MLLMs) offers an unprecedented opportunity to evolve and empower game agents with human-like decision-making capabilities in complex computer game environments. This paper provides a comprehensive overview of LLM-based game agents from a holistic viewpoint. First, we introduce the conceptual architecture of LLM-based game agents, centered around six essential functional components: perception, memory, thinking, role-playing, action, and learning. Second, we survey existing representative LLM-based game agents documented in the literature with respect to methodologies and adaptation agility across six genres of games, including adventure, communication, competition, cooperation, simulation, and crafting & exploration games. Finally, we present an outlook of future research and development directions in this burgeoning field.

Well-trained deep neural networks (DNNs) treat all test samples equally during prediction. Adaptive DNN inference with early exiting leverages the observation that some test examples can be easier to predict than others. This paper presents EENet, a novel early-exiting scheduling framework for multi-exit DNN models. Instead of having every sample go through all DNN layers during prediction, EENet learns an early exit scheduler, which can intelligently terminate the inference earlier for certain predictions, which the model has high confidence of early exit. As opposed to previous early-exiting solutions with heuristics-based methods, our EENet framework optimizes an early-exiting policy to maximize model accuracy while satisfying the given per-sample average inference budget. Extensive experiments are conducted on four computer vision datasets (CIFAR-10, CIFAR-100, ImageNet, Cityscapes) and two NLP datasets (SST-2, AgNews). The results demonstrate that the adaptive inference by EENet can outperform the representative existing early exit techniques. We also perform a detailed visualization analysis of the comparison results to interpret the benefits of EENet.