Afterburner: Reinforcement Learning Facilitates Self-Improving Code Efficiency Optimization

Large Language Models (LLMs) generate functionally correct solutions but often fall short in code efficiency, a critical bottleneck for real-world deployment. In this paper, we introduce a novel test-time iterative optimization framework to address this, employing a closed-loop system where LLMs iteratively refine code based on empirical performance feedback from an execution sandbox. We explore three training strategies: Supervised Fine-Tuning (SFT), Direct Preference Optimization (DPO), and Group Relative Policy Optimization~(GRPO). Experiments on our Venus dataset and the APPS benchmark show that SFT and DPO rapidly saturate in efficiency gains. In contrast, GRPO, using reinforcement learning (RL) with execution feedback, continuously optimizes code performance, significantly boosting both pass@1 (from 47% to 62%) and the likelihood of outperforming human submissions in efficiency (from 31% to 45%). Our work demonstrates effective test-time code efficiency improvement and critically reveals the power of RL in teaching LLMs to truly self-improve code efficiency.

Sparse-to-Dense: A Free Lunch for Lossless Acceleration of Video Understanding in LLMs

Due to the auto-regressive nature of current video large language models (Video-LLMs), the inference latency increases as the input sequence length grows, posing challenges for the efficient processing of video sequences that are usually very long. We observe that during decoding, the attention scores of most tokens in Video-LLMs tend to be sparse and concentrated, with only certain tokens requiring comprehensive full attention. Based on this insight, we introduce Sparse-to-Dense (StD), a novel decoding strategy that integrates two distinct modules: one leveraging sparse top-K attention and the other employing dense full attention. These modules collaborate to accelerate Video-LLMs without loss. The fast (sparse) model speculatively decodes multiple tokens, while the slow (dense) model verifies them in parallel. StD is a tuning-free, plug-and-play solution that achieves up to a 1.94$\times$ walltime speedup in video processing. It maintains model performance while enabling a seamless transition from a standard Video-LLM to a sparse Video-LLM with minimal code modifications.

DiffusionRL: Efficient Training of Diffusion Policies for Robotic Grasping Using RL-Adapted Large-Scale Datasets

Diffusion models have been successfully applied in areas such as image, video, and audio generation. Recent works show their promise for sequential decision-making and dexterous manipulation, leveraging their ability to model complex action distributions. However, challenges persist due to the data limitations and scenario-specific adaptation needs. In this paper, we address these challenges by proposing an optimized approach to training diffusion policies using large, pre-built datasets that are enhanced using Reinforcement Learning (RL). Our end-to-end pipeline leverages RL-based enhancement of the DexGraspNet dataset, lightweight diffusion policy training on a dexterous manipulation task for a five-fingered robotic hand, and a pose sampling algorithm for validation. The pipeline achieved a high success rate of 80% for three DexGraspNet objects. By eliminating manual data collection, our approach lowers barriers to adopting diffusion models in robotics, enhancing generalization and robustness for real-world applications.

SWE-Dev: Evaluating and Training Autonomous Feature-Driven Software Development

Large Language Models (LLMs) have shown strong capability in diverse software engineering tasks, e.g. code completion, bug fixing, and document generation. However, feature-driven development (FDD), a highly prevalent real-world task that involves developing new functionalities for large, existing codebases, remains underexplored. We therefore introduce SWE-Dev, the first large-scale dataset (with 14,000 training and 500 test samples) designed to evaluate and train autonomous coding systems on real-world feature development tasks. To ensure verifiable and diverse training, SWE-Dev uniquely provides all instances with a runnable environment and its developer-authored executable unit tests. This collection not only provides high-quality data for Supervised Fine-Tuning (SFT), but also enables Reinforcement Learning (RL) by delivering accurate reward signals from executable unit tests. Our extensive evaluations on SWE-Dev, covering 17 chatbot LLMs, 10 reasoning models, and 10 Multi-Agent Systems (MAS), reveal that FDD is a profoundly challenging frontier for current AI (e.g., Claude-3.7-Sonnet achieves only 22.45\% Pass@3 on the hard test split). Crucially, we demonstrate that SWE-Dev serves as an effective platform for model improvement: fine-tuning on training set enabled a 7B model comparable to GPT-4o on \textit{hard} split, underscoring the value of its high-quality training data. Code is available here \href{https://github.com/justLittleWhite/SWE-Dev}{https://github.com/justLittleWhite/SWE-Dev}.

General-Reasoner: Advancing LLM Reasoning Across All Domains

Reinforcement learning (RL) has recently demonstrated strong potential in enhancing the reasoning capabilities of large language models (LLMs). Particularly, the "Zero" reinforcement learning introduced by Deepseek-R1-Zero, enables direct RL training of base LLMs without relying on an intermediate supervised fine-tuning stage. Despite these advancements, current works for LLM reasoning mainly focus on mathematical and coding domains, largely due to data abundance and the ease of answer verification. This limits the applicability and generalization of such models to broader domains, where questions often have diverse answer representations, and data is more scarce. In this paper, we propose General-Reasoner, a novel training paradigm designed to enhance LLM reasoning capabilities across diverse domains. Our key contributions include: (1) constructing a large-scale, high-quality dataset of questions with verifiable answers curated by web crawling, covering a wide range of disciplines; and (2) developing a generative model-based answer verifier, which replaces traditional rule-based verification with the capability of chain-of-thought and context-awareness. We train a series of models and evaluate them on a wide range of datasets covering wide domains like physics, chemistry, finance, electronics etc. Our comprehensive evaluation across these 12 benchmarks (e.g. MMLU-Pro, GPQA, SuperGPQA, TheoremQA, BBEH and MATH AMC) demonstrates that General-Reasoner outperforms existing baseline methods, achieving robust and generalizable reasoning performance while maintaining superior effectiveness in mathematical reasoning tasks.

Impact of Insufficient CP on Sensing Performance in OFDM-ISAC Systems

Orthogonal frequency-division multiplexing (OFDM) is widely considered a leading waveform candidate for integrated sensing and communication (ISAC) in 6G networks. However, the cyclic prefix (CP) used to mitigate multipath effects in communication systems also limits the maximum sensing range. Target echoes arriving beyond the CP length cause inter-symbol interference (ISI) and inter-carrier interference (ICI), which degrade the mainlobe level and raise sidelobe levels in the range-Doppler map (RDM). This paper presents a unified analytical framework to characterize the ISI and ICI caused by an insufficient CP length in multi-target scenarios. For the first time, we derive closed-form expressions for the second-order moments of the RDM under both matched filtering (MF) and reciprocal filtering (RF) processing with insufficient CP length. These expressions quantify the effects of CP length, symbol constellation, and inter-target interference (ITI) on the mainlobe and sidelobe levels. Based on these results, we further derive explicit formulas for the peak sidelobe level ratio (PSLR) and integrated sidelobe level ratio (ISLR) of the RDM, revealing a fundamental trade-off between noise amplification in RF and ITI in MF. Numerical results validate our theoretical derivations and illustrate the critical impact of insufficient CP length on sensing performance in OFDM-ISAC systems.

Sensing-Oriented Adaptive Resource Allocation Designs for OFDM-ISAC Systems

Orthogonal frequency division multiplexing - integrated sensing and communication (OFDM-ISAC) has emerged as a key enabler for future wireless networks, leveraging the widely adopted OFDM waveform to seamlessly integrate wireless communication and radar sensing within a unified framework. In this paper, we propose adaptive resource allocation strategies for OFDM-ISAC systems to achieve optimal trade-offs between diverse sensing requirements and communication quality-of-service (QoS). We first develop a comprehensive resource allocation framework for OFDM-ISAC systems, deriving closed-form expressions for key sensing performance metrics, including delay resolution, Doppler resolution, delay-Doppler peak sidelobe level (PSL), and received signal-to-noise ratio (SNR). Building on this theoretical foundation, we introduce two novel resource allocation algorithms tailored to distinct sensing objectives. The resolution-oriented algorithm aims to maximize the weighted delay-Doppler resolution while satisfying constraints on PSL, sensing SNR, communication sum-rate, and transmit power. The sidelobe-oriented algorithm focuses on minimizing delay-Doppler PSL while satisfying resolution, SNR, and communication constraints. To efficiently solve the resulting non-convex optimization problems, we develop two adaptive resource allocation algorithms based on Dinkelbach's transform and majorization-minimization (MM). Extensive simulations validate the effectiveness of the proposed sensing-oriented adaptive resource allocation strategies in enhancing resolution and sidelobe suppression. Remarkably, these strategies achieve sensing performance nearly identical to that of a radar-only scheme, which dedicates all resources to sensing. These results highlight the superior performance of the proposed methods in optimizing the trade-off between sensing and communication objectives within OFDM-ISAC systems.

SimpleRL-Zoo: Investigating and Taming Zero Reinforcement Learning for Open Base Models in the Wild

DeepSeek-R1 has shown that long chain-of-thought (CoT) reasoning can naturally emerge through a simple reinforcement learning (RL) framework with rule-based rewards, where the training may directly start from the base models-a paradigm referred to as zero RL training. Most recent efforts to reproduce zero RL training have primarily focused on the Qwen2.5 model series, which may not be representative as we find the base models already exhibit strong instruction-following and self-reflection abilities. In this work, we investigate zero RL training across 10 diverse base models, spanning different families and sizes including LLama3-8B, Mistral-7B/24B, DeepSeek-Math-7B, Qwen2.5-math-7B, and all Qwen2.5 models from 0.5B to 32B. Leveraging several key design strategies-such as adjusting format reward and controlling query difficulty-we achieve substantial improvements in both reasoning accuracy and response length across most settings. However, by carefully monitoring the training dynamics, we observe that different base models exhibit distinct patterns during training. For instance, the increased response length does not always correlate with the emergence of certain cognitive behaviors such as verification (i.e., the "aha moment"). Notably, we observe the "aha moment" for the first time in small models not from the Qwen family. We share the key designs that enable successful zero RL training, along with our findings and practices. To facilitate further research, we open-source the code, models, and analysis tools.

CONTHER: Human-Like Contextual Robot Learning via Hindsight Experience Replay and Transformers without Expert Demonstrations

This paper presents CONTHER, a novel reinforcement learning algorithm designed to efficiently and rapidly train robotic agents for goal-oriented manipulation tasks and obstacle avoidance. The algorithm uses a modified replay buffer inspired by the Hindsight Experience Replay (HER) approach to artificially populate experience with successful trajectories, effectively addressing the problem of sparse reward scenarios and eliminating the need to manually collect expert demonstrations. The developed algorithm proposes a Transformer-based architecture to incorporate the context of previous states, allowing the agent to perform a deeper analysis and make decisions in a manner more akin to human learning. The effectiveness of the built-in replay buffer, which acts as an "internal demonstrator", is twofold: it accelerates learning and allows the algorithm to adapt to different tasks. Empirical data confirm the superiority of the algorithm by an average of 38.46% over other considered methods, and the most successful baseline by 28.21%, showing higher success rates and faster convergence in the point-reaching task. Since the control is performed through the robot's joints, the algorithm facilitates potential adaptation to a real robot system and construction of an obstacle avoidance task. Therefore, the algorithm has also been tested on tasks requiring following a complex dynamic trajectory and obstacle avoidance. The design of the algorithm ensures its applicability to a wide range of goal-oriented tasks, making it an easily integrated solution for real-world robotics applications.

SkyLadder: Better and Faster Pretraining via Context Window Scheduling

Recent advancements in LLM pretraining have featured ever-expanding context windows to process longer sequences. However, our pilot study reveals that models pretrained with shorter context windows consistently outperform their long-context counterparts under a fixed token budget. This finding motivates us to explore an optimal context window scheduling strategy to better balance long-context capability with pretraining efficiency. To this end, we propose SkyLadder, a simple yet effective approach that implements a short-to-long context window transition. SkyLadder preserves strong standard benchmark performance, while matching or exceeding baseline results on long context tasks. Through extensive experiments, we pre-train 1B-parameter models (up to 32K context) and 3B-parameter models (8K context) on 100B tokens, demonstrating that SkyLadder yields consistent gains of up to 3.7% on common benchmarks, while achieving up to 22% faster training speeds compared to baselines. The code is at https://github.com/sail-sg/SkyLadder.