We propose ADAPT, a meta-learning algorithm that \emph{learns} task sampling proportions under an explicit token budget for multi-task instruction tuning. Instead of fixing task weights by hand, \adapt{} maintains a continuous distribution over tasks and updates it via meta-gradients of a smooth worst-case validation objective, inducing an adaptive curriculum that allocates more tokens to useful tasks while avoiding collapse. We instantiate ADAPT on three $\sim$1B-parameter open-weight LLMs (Gemma-3-1B, LLaMA-3.2-1B, Qwen-0.6B), training on 20 Natural Instructions task types under budgets of $1\%$, $5\%$, and $10\%$ of the available supervised tokens, and compare against strong supervised fine-tuning baselines with uniform and size-proportional mixing. We conduct evaluations on 11 out-of-domain benchmarks spanning reasoning, reading comprehension, code generation, and instruction following, we find that ADAPT matches or slightly improves average downstream performance relative to the best static mixture, while using fewer effective training tokens and reallocating budget toward harder, benchmark-aligned tasks.

Creative Thinking(TTCW) benchmark introduced in Chakrabarty et al. (2024), Rigorous, standardized evaluation has repeatedly catalyzed progress in machine learning, from ImageNetRussakovsky et al. (2015) and GLUEWang et al. (2019), driving leaps in the fields of computer vision and Natural Language Processing, respectively. The same effect is evident in objective math reasoning, where benchmarks like GSM8KCobbe et al. (2021), together with RL-trained reasoning models such as OpenAI's o1OpenAI et al. (2024) and DeepSeek-R1DeepSeek-AI Models(LLM) as a judge prefer their own generations making them unreliable. As shown in Chakrabarty et al. (2024) and Table 12 and Table 2, even Specifically, when the model is "surprised" by an expert's explanation, it signals a mismatch between the LLM's prior belief and the expert's The intuition behind predicting the annotator is that the model can learn which annotator caused the belief shift, allowing it to calibrate the curiosity signal for each annotator individually, thereby improving personalization. In our experiments, we establish a baseline using an SFT model that predicts annotators' binary More details about the results can be found in Fig 4.Figure 1: Overview of Architecture during training for Curiosity Driven LLM-as-a-judgeFigure 2: Overview of Architecture during inference for Curiosity Driven LLM-as-a-judge 2 (a) Baseline without using explanations (b) Baseline using explanations TTCW dataset Chakrabarty et al. (2024), which is based on the Torrance Test of Creative Thinking Torrance (1966) but adapted for LLMs. All the distinct dimensions in the TTCW dataset are mentioned in Appendix A.1.

Graphical user interface visual grounding (GUI-VG), a core capability for GUI agents, has primarily relied on supervised fine-tuning (SFT) of multimodal large language models (MLLMs), which demands extensive data curation and significant training costs. However, as MLLMs continue to advance and even cover GUI domains during pretraining, the necessity of exhaustive SFT post-training becomes increasingly questionable. Meanwhile, recent successes of rule-based reinforcement fine-tuning (RFT) suggest a more efficient alternative. Despite this promise, the optimal manner of applying RFT for GUI-VG remains unexplored. To bridge this gap, we introduce GuirlVG, a reinforcement learning-based GUI-VG method built on a systematic empirical study and a novel stabilization technique. We find that naive application of RFT underperforms the SFT baseline, motivating a deeper exploration. First, we decompose RFT into its core components and analyze the optimal formulation of each. Second, we propose a novel Adversarial KL Factor that dynamically stabilizes training to mitigate reward over-optimization. Third, we further explore the training configurations of RFT to enhance effectiveness. Extensive experiments show that GuirlVG, with only 5.2K training samples, outperforms SFT methods trained on over 10M samples, achieving a 7.7% improvement on ScreenSpot, a 17.2% improvement on ScreenSpotPro, and 91.9% accuracy on ScreenSpotV2.

Reinforcement Learning from Verifiable Rewards (RLVR) has been widely adopted as the de facto method for enhancing the reasoning capabilities of large language models and has demonstrated notable success in verifiable domains like math and competitive programming tasks. However, the efficacy of RLVR diminishes significantly when applied to agentic environments. These settings, characterized by multi-step, complex problem solving, lead to high failure rates even for frontier LLMs, as the reward landscape is too sparse for effective model training via conventional RLVR. In this work, we introduce Agent-RLVR, a framework that makes RLVR effective in challenging agentic settings, with an initial focus on software engineering tasks. Inspired by human pedagogy, Agent-RLVR introduces agent guidance, a mechanism that actively steers the agent towards successful trajectories by leveraging diverse informational cues. These cues, ranging from high-level strategic plans to dynamic feedback on the agent's errors and environmental interactions, emulate a teacher's guidance, enabling the agent to navigate difficult solution spaces and promotes active self-improvement via additional environment exploration. In the Agent-RLVR training loop, agents first attempt to solve tasks to produce initial trajectories, which are then validated by unit tests and supplemented with agent guidance. Agents then reattempt with guidance, and the agent policy is updated with RLVR based on the rewards of these guided trajectories. Agent-RLVR elevates the pass@1 performance of Qwen-2.5-72B-Instruct from 9.4% to 22.4% on SWE-Bench Verified. We find that our guidance-augmented RLVR data is additionally useful for test-time reward model training, shown by further boosting pass@1 to 27.8%. Agent-RLVR lays the groundwork for training agents with RLVR in complex, real-world environments where conventional RL methods struggle.

Radiologists play a crucial role by translating medical images into medical reports. However, the field faces staffing shortages and increasing workloads. While automated approaches using vision-language models (VLMs) show promise as assistants, they require exceptionally high accuracy. Most current VLMs in radiology rely solely on supervised fine-tuning (SFT). Meanwhile, in the general domain, additional preference fine-tuning has become standard practice. The challenge in radiology lies in the prohibitive cost of obtaining radiologist feedback. We propose a scalable automated preference alignment technique for VLMs in radiology, focusing on chest X-ray (CXR) report generation. Our method leverages publicly available datasets with an LLM-as-a-Judge mechanism, eliminating the need for additional expert radiologist feedback. We evaluate and benchmark five direct alignment algorithms (DAAs). Our results show up to a 57.4% improvement in average GREEN scores, a LLM-based metric for evaluating CXR reports, and a 9.2% increase in an average across six metrics (domain specific and general), compared to the SFT baseline. We study reward overoptimization via length exploitation, with reports lengthening by up to 3.2x. To assess a potential alignment tax, we benchmark on six additional diverse tasks, finding no significant degradations. A reader study involving four board-certified radiologists indicates win rates of up to 0.62 over the SFT baseline, while significantly penalizing verbosity. Our analysis provides actionable insights for the development of VLMs in high-stakes fields like radiology.