Finetuning open-weight Large Language Models (LLMs) is standard practice for achieving task-specific performance improvements. Until now, finetuning has been regarded as a controlled and secure process in which training on benign datasets leads to predictable behaviors. In this paper, we demonstrate, for the first time, that an adversary can create compromised LLMs that are performant and benign, yet exhibit adversarial behaviors once finetuned by downstream users. To this end, we propose an attack, FAB (Finetuning-activated Adversarial Behaviors), which compromises an LLM via meta-learning techniques that simulate downstream finetuning, explicitly optimizing for the emergence of adversarial behaviors in the finetuned models. At the same time, the compromised LLM is regularized to retain general capabilities and to exhibit no adversarial behaviors prior to finetuning. As a result, when users finetune (e.g., instruction-tuning, distillation, DPO) the seemingly benign model on their own datasets, they unknowingly trigger its dormant adversarial behavior. We experimentally demonstrate the effectiveness of FAB across multiple LLMs and three commonly considered target behaviors: unsolicited advertising, jailbreakability, and over-refusal. We show that FAB-triggers are robust to various finetuning choices made by the user (e.g., dataset, number of steps, scheduler, post-training algorithm). Our findings challenge prevailing assumptions on the security of finetuning, revealing a critical attack vector.

While reinforcement learning (RL) has demonstrated remarkable success in enhancing large language models (LLMs), it has primarily focused on single-turn tasks such as solving math problems. Training effective web agents for multi-turn interactions remains challenging due to the complexity of long-horizon decision-making across dynamic web interfaces. In this work, we present WebAgent-R1, a simple yet effective end-to-end multi-turn RL framework for training web agents. It learns directly from online interactions with web environments by asynchronously generating diverse trajectories, entirely guided by binary rewards depending on task success. Experiments on the WebArena-Lite benchmark demonstrate the effectiveness of WebAgent-R1, boosting the task success rate of Qwen-2.5-3B from 6.1% to 33.9% and Llama-3.1-8B from 8.5% to 44.8%, significantly outperforming existing state-of-the-art methods and strong proprietary models such as OpenAI o3. In-depth analyses reveal the effectiveness of the thinking-based prompting strategy and test-time scaling through increased interactions for web tasks. We further investigate different RL initialization policies by introducing two variants, namely WebAgent-R1-Zero and WebAgent-R1-CoT, which highlight the importance of the warm-up training stage (i.e., behavior cloning) and provide insights on incorporating long chain-of-thought (CoT) reasoning in web agents.

Despite substantial advancements in aligning large language models (LLMs) with human values, current safety mechanisms remain susceptible to jailbreak attacks. We hypothesize that this vulnerability stems from distributional discrepancies between alignment-oriented prompts and malicious prompts. To investigate this, we introduce LogiBreak, a novel and universal black-box jailbreak method that leverages logical expression translation to circumvent LLM safety systems. By converting harmful natural language prompts into formal logical expressions, LogiBreak exploits the distributional gap between alignment data and logic-based inputs, preserving the underlying semantic intent and readability while evading safety constraints. We evaluate LogiBreak on a multilingual jailbreak dataset spanning three languages, demonstrating its effectiveness across various evaluation settings and linguistic contexts.

Large reasoning models (LRMs) are proficient at generating explicit, step-by-step reasoning sequences before producing final answers. However, such detailed reasoning can introduce substantial computational overhead and latency, particularly for simple problems. To address this over-thinking problem, we explore how to equip LRMs with adaptive thinking capabilities: enabling them to dynamically decide whether or not to engage in explicit reasoning based on problem complexity. Building on R1-style distilled models, we observe that inserting a simple ellipsis ("...") into the prompt can stochastically trigger either a thinking or no-thinking mode, revealing a latent controllability in the reasoning behavior. Leveraging this property, we propose AutoThink, a multi-stage reinforcement learning (RL) framework that progressively optimizes reasoning policies via stage-wise reward shaping. AutoThink learns to invoke explicit reasoning only when necessary, while defaulting to succinct responses for simpler tasks. Experiments on five mainstream mathematical benchmarks demonstrate that AutoThink achieves favorable accuracy-efficiency trade-offs compared to recent prompting and RL-based pruning methods. It can be seamlessly integrated into any R1-style model, including both distilled and further fine-tuned variants. Notably, AutoThink improves relative accuracy by 6.4 percent while reducing token usage by 52 percent on DeepSeek-R1-Distill-Qwen-1.5B, establishing a scalable and adaptive reasoning paradigm for LRMs. Project Page: https://github.com/ScienceOne-AI/AutoThink.

To perform few-shot learning, language models extract signals from a few input-label pairs, aggregate these into a learned prediction rule, and apply this rule to new inputs. How is this implemented in the forward pass of modern transformer models? To explore this question, we study a structured family of few-shot learning tasks for which the true prediction rule is to add an integer $k$ to the input. We introduce a novel optimization method that localizes the model's few-shot ability to only a few attention heads. We then perform an in-depth analysis of individual heads, via dimensionality reduction and decomposition. As an example, on Llama-3-8B-instruct, we reduce its mechanism on our tasks to just three attention heads with six-dimensional subspaces, where four dimensions track the unit digit with trigonometric functions at periods $2$, $5$, and $10$, and two dimensions track magnitude with low-frequency components. To deepen our understanding of the mechanism, we also derive a mathematical identity relating ``aggregation'' and ``extraction'' subspaces for attention heads, allowing us to track the flow of information from individual examples to a final aggregated concept. Using this, we identify a self-correction mechanism where mistakes learned from earlier demonstrations are suppressed by later demonstrations. Our results demonstrate how tracking low-dimensional subspaces of localized heads across a forward pass can provide insight into fine-grained computational structures in language models.

Modern machine learning (ML) training workloads place substantial demands on both computational and communication resources. Consequently, accurate performance estimation has become increasingly critical for guiding system design decisions, such as the selection of parallelization strategies, cluster configurations, and hardware provisioning. Existing simulation-based performance estimation requires reimplementing the ML framework in a simulator, which demands significant manual effort and is hard to maintain as ML frameworks evolve rapidly. This paper introduces Phantora, a hybrid GPU cluster simulator designed for performance estimation of ML training workloads. Phantora executes unmodified ML frameworks as is within a distributed, containerized environment. Each container emulates the behavior of a GPU server in a large-scale cluster, while Phantora intercepts and simulates GPU- and communication-related operations to provide high-fidelity performance estimation. We call this approach hybrid simulation of ML systems, in contrast to traditional methods that simulate static workloads. The primary advantage of hybrid simulation is that it allows direct reuse of ML framework source code in simulation, avoiding the need for reimplementation. Our evaluation shows that Phantora provides accuracy comparable to static workload simulation while supporting three state-of-the-art LLM training frameworks out-of-the-box. In addition, Phantora operates on a single GPU, eliminating the need for the resource-intensive trace collection and workload extraction steps required by traditional trace-based simulators. Phantora is open-sourced at https://github.com/QDelta/Phantora.

Generative models have achieved remarkable success across various applications, driving the demand for multi-GPU computing. Inter-GPU communication becomes a bottleneck in multi-GPU computing systems, particularly on consumer-grade GPUs. By exploiting concurrent hardware execution, overlapping computation and communication latency becomes an effective technique for mitigating the communication overhead. We identify that an efficient and adaptable overlapping design should satisfy (1) tile-wise overlapping to maximize the overlapping opportunity, (2) interference-free computation to maintain the original computational performance, and (3) communication agnosticism to reduce the development burden against varying communication primitives. Nevertheless, current designs fail to simultaneously optimize for all of those features. To address the issue, we propose FlashOverlap, which utilizes a novel signaling mechanism: when part of the output finishes, the computation kernel sends a signal to trigger the communication of that part, while continuing the computation of the remaining part (interference-free computation). Consequently, the communication of the finished part and the computation of the remaining part can be overlapped. On top of the signaling mechanism, FlashOverlap comprises two key components: (1) the determination of the signaling timing to boost the overlap efficiency (tile-wise overlapping), and (2) a pre-communication reordering to create the contiguous address for finished data, enabling communication by simply calling NCCL APIs (communication agnosticism), and a post-communication reordering to correct the data order. Experiments show that FlashOverlap achieves up to 1.65x speedup through overlap, outperforming existing works in most cases. Code is available at https://github.com/infinigence/FlashOverlap.

Hallucinations pose a significant obstacle to the reliability and widespread adoption of language models, yet their accurate measurement remains a persistent challenge. While many task- and domain-specific metrics have been proposed to assess faithfulness and factuality concerns, the robustness and generalization of these metrics are still untested. In this paper, we conduct a large-scale empirical evaluation of 6 diverse sets of hallucination detection metrics across 4 datasets, 37 language models from 5 families, and 5 decoding methods. Our extensive investigation reveals concerning gaps in current hallucination evaluation: metrics often fail to align with human judgments, take an overtly myopic view of the problem, and show inconsistent gains with parameter scaling. Encouragingly, LLM-based evaluation, particularly with GPT-4, yields the best overall results, and mode-seeking decoding methods seem to reduce hallucinations, especially in knowledge-grounded settings. These findings underscore the need for more robust metrics to understand and quantify hallucinations, and better strategies to mitigate them.

The growing capabilities of Artificial Intelligence (AI), particularly Large Language Models (LLMs), prompt a reassessment of the interaction mechanisms between users and their devices. Currently, users are required to use a set of high-level applications to achieve their desired results. However, the advent of AI may signal a shift in this regard, as its capabilities have generated novel prospects for user-provided intent resolution through the deployment of model-generated code. This development represents a significant progression in the realm of hybrid workflows, where human and artificial intelligence collaborate to address user intentions, with the former responsible for defining these intentions and the latter for implementing the solutions to address them. In this paper, we investigate the feasibility of generating and executing workflows through code generation that results from prompting an LLM with a concrete user intention, and a simplified application programming interface for a GUI-less operating system. We provide an in-depth analysis and comparison of various user intentions, the resulting code, and its execution. The findings demonstrate the general feasibility of our approach and that the employed LLM, GPT -4o-mini, exhibits remarkable proficiency in the generation of code-oriented workflows in accordance with provided user intentions.

The specialized vocabulary and nuanced concepts of the telecommunications industry pose persistent challenges for standard Natural Language Processing (NLP) models. Generic embedding models often struggle to represent telecom-specific semantics, limiting their utility in retrieval and downstream tasks. We present T-VEC (Telecom Vectorization Model), a domain-adapted embedding model fine-tuned from the gte-Qwen2-1.5B-instruct backbone using a triplet loss objective. Fine-tuning was performed on T-Embed, a high-quality, large-scale dataset covering diverse telecom concepts, standards, and operational scenarios. Although T-Embed contains some proprietary material and cannot be fully released, we open source 75% of the dataset to support continued research in domain-specific representation learning. On a custom benchmark comprising 1500 query-passage pairs from IETF RFCs and vendor manuals, T-VEC surpasses MPNet, BGE, Jina and E5, demonstrating superior domain grounding and semantic precision in telecom-specific retrieval. Embedding visualizations further showcase tight clustering of telecom-relevant concepts. We release T-VEC and its tokenizer to support semantically faithful NLP applications within the telecom domain.