How Far Are We from Optimal Reasoning Efficiency?

Large Reasoning Models (LRMs) demonstrate remarkable problem-solving capabilities through extended Chain-of-Thought (CoT) reasoning but often produce excessively verbose and redundant reasoning traces. This inefficiency incurs high inference costs and limits practical deployment. While existing fine-tuning methods aim to improve reasoning efficiency, assessing their efficiency gains remains challenging due to inconsistent evaluations. In this work, we introduce the reasoning efficiency frontiers, empirical upper bounds derived from fine-tuning base LRMs across diverse approaches and training configurations. Based on these frontiers, we propose the Reasoning Efficiency Gap (REG), a unified metric quantifying deviations of any fine-tuned LRMs from these frontiers. Systematic evaluation on challenging mathematical benchmarks reveals significant gaps in current methods: they either sacrifice accuracy for short length or still remain inefficient under tight token budgets. To reduce the efficiency gap, we propose REO-RL, a class of Reinforcement Learning algorithms that minimizes REG by targeting a sparse set of token budgets. Leveraging numerical integration over strategically selected budgets, REO-RL approximates the full efficiency objective with low error using a small set of token budgets. Through systematic benchmarking, we demonstrate that our efficiency metric, REG, effectively captures the accuracy-length trade-off, with low-REG methods reducing length while maintaining accuracy. Our approach, REO-RL, consistently reduces REG by >=50 across all evaluated LRMs and matching Qwen3-4B/8B efficiency frontiers under a 16K token budget with minimal accuracy loss. Ablation studies confirm the effectiveness of our exponential token budget strategy. Finally, our findings highlight that fine-tuning LRMs to perfectly align with the efficiency frontiers remains an open challenge.

AReaL: A Large-Scale Asynchronous Reinforcement Learning System for Language Reasoning

Reinforcement learning (RL) has become a trending paradigm for training large language models (LLMs), particularly for reasoning tasks. Effective RL for LLMs requires massive parallelization and poses an urgent need for efficient training systems. Most existing large-scale RL systems for LLMs are synchronous by alternating generation and training in a batch setting, where the rollouts in each training batch are generated by the same (or latest) model. This stabilizes RL training but suffers from severe system-level inefficiency. Generation must wait until the longest output in the batch is completed before model update, resulting in GPU underutilization. We present AReaL, a \emph{fully asynchronous} RL system that completely decouples generation from training. Rollout workers in AReaL continuously generate new outputs without waiting, while training workers update the model whenever a batch of data is collected. AReaL also incorporates a collection of system-level optimizations, leading to substantially higher GPU utilization. To stabilize RL training, AReaL balances the workload of rollout and training workers to control data staleness, and adopts a staleness-enhanced PPO variant to better handle outdated training samples. Extensive experiments on math and code reasoning benchmarks show that AReaL achieves \textbf{up to 2.57$\times$ training speedup} compared to the best synchronous systems with the same number of GPUs and matched or even improved final performance. The code of AReaL is available at https://github.com/inclusionAI/AReaL/.

On Designing Effective RL Reward at Training Time for LLM Reasoning

Reward models have been increasingly critical for improving the reasoning capability of LLMs. Existing research has shown that a well-trained reward model can substantially improve model performances at inference time via search. However, the potential of reward models during RL training time still remains largely under-explored. It is currently unclear whether these reward models can provide additional training signals to enhance the reasoning capabilities of LLMs in RL training that uses sparse success rewards, which verify the correctness of solutions. In this work, we evaluate popular reward models for RL training, including the Outcome-supervised Reward Model (ORM) and the Process-supervised Reward Model (PRM), and train a collection of LLMs for math problems using RL by combining these learned rewards with success rewards. Surprisingly, even though these learned reward models have strong inference-time performances, they may NOT help or even hurt RL training, producing worse performances than LLMs trained with the success reward only. Our analysis reveals that an LLM can receive high rewards from some of these reward models by repeating correct but unnecessary reasoning steps, leading to a severe reward hacking issue. Therefore, we introduce two novel reward refinement techniques, including Clipping and Delta. The key idea is to ensure the accumulative reward of any reasoning trajectory is upper-bounded to keep a learned reward model effective without being exploited. We evaluate our techniques with multiple reward models over a set of 1.5B and 7B LLMs on MATH and GSM8K benchmarks and demonstrate that with a carefully designed reward function, RL training without any additional supervised tuning can improve all the evaluated LLMs, including the state-of-the-art 7B LLM Qwen2.5-Math-7B-Instruct on MATH and GSM8K benchmarks.

Is DPO Superior to PPO for LLM Alignment? A Comprehensive Study

Reinforcement Learning from Human Feedback (RLHF) is currently the most widely used method to align large language models (LLMs) with human preferences. Existing RLHF methods can be roughly categorized as either reward-based or reward-free. Novel applications such as ChatGPT and Claude leverage reward-based methods that first learn a reward model and apply actor-critic algorithms, such as Proximal Policy Optimization (PPO). However, in academic benchmarks, state-of-the-art results are often achieved via reward-free methods, such as Direct Preference Optimization (DPO). Is DPO truly superior to PPO? Why does PPO perform poorly on these benchmarks? In this paper, we first conduct both theoretical and empirical studies on the algorithmic properties of DPO and show that DPO may have fundamental limitations. Moreover, we also comprehensively examine PPO and reveal the key factors for the best performances of PPO in fine-tuning LLMs. Finally, we benchmark DPO and PPO across various a collection of RLHF testbeds, ranging from dialogue to code generation. Experiment results demonstrate that PPO is able to surpass other alignment methods in all cases and achieve state-of-the-art results in challenging code competitions.

Language-Guided Generation of Physically Realistic Robot Motion and Control

We aim to control a robot to physically behave in the real world following any high-level language command like "cartwheel" or "kick. " Although human motion datasets exist, this task remains particularly challenging since generative models can produce physically unrealistic motions, which will be more severe for robots due to different body structures and physical properties. In addition, to control a physical robot to perform a desired motion, a control policy must be learned. We develop LAnguage-Guided mOtion cONtrol (LAGOON), a multi-phase method to generate physically realistic robot motions under language commands. LAGOON first leverages a pre-trained model to generate human motion from a language command. Then an RL phase is adopted to train a control policy in simulation to mimic the generated human motion. Finally, with domain randomization, we show that our learned policy can be successfully deployed to a quadrupedal robot, leading to a robot dog that can stand up and wave its front legs in the real world to mimic the behavior of a hand-waving human.

Native Chinese Reader: A Dataset Towards Native-Level Chinese Machine Reading Comprehension

We present Native Chinese Reader (NCR), a new machine reading comprehension (MRC) dataset with particularly long articles in both modern and classical Chinese. NCR is collected from the exam questions for the Chinese course in China's high schools, which are designed to evaluate the language proficiency of native Chinese youth. Existing Chinese MRC datasets are either domain-specific or focusing on short contexts of a few hundreds of characters in modern Chinese only. By contrast, NCR contains 8390 documents with an average length of 1024 characters covering a wide range of Chinese writing styles, including modern articles, classical literature and classical poetry. A total of 20477 questions on these documents also require strong reasoning abilities and common sense to figure out the correct answers. We implemented multiple baseline models using popular Chinese pre-trained models and additionally launched an online competition using our dataset to examine the limit of current methods. The best model achieves 59% test accuracy while human evaluation shows an average accuracy of 79%, which indicates a significant performance gap between current MRC models and native Chinese speakers. We release the dataset at https://sites.google.com/view/native-chinese-reader/.

A Benchmark for Low-Switching-Cost Reinforcement Learning

A ubiquitous requirement in many practical reinforcement learning (RL) applications, including medical treatment, recommendation system, education and robotics, is that the deployed policy that actually interacts with the environment cannot change frequently. Such an RL setting is called low-switching-cost RL, i.e., achieving the highest reward while reducing the number of policy switches during training. Despite the recent trend of theoretical studies aiming to design provably efficient RL algorithms with low switching costs, none of the existing approaches have been thoroughly evaluated in popular RL testbeds. In this paper, we systematically studied a wide collection of policy-switching approaches, including theoretically guided criteria, policy-difference-based methods, and non-adaptive baselines. Through extensive experiments on a medical treatment environment, the Atari games, and robotic control tasks, we present the first empirical benchmark for low-switching-cost RL and report novel findings on how to decrease the switching cost while maintain a similar sample efficiency to the case without the low-switching-cost constraint. We hope this benchmark could serve as a starting point for developing more practically effective low-switching-cost RL algorithms. We release our code and complete results in https://sites.google.com/view/low-switching-cost-rl.

PhyloTransformer: A Discriminative Model for Mutation Prediction Based on a Multi-head Self-attention Mechanism

Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has caused an ongoing pandemic infecting 219 million people as of 10/19/21, with a 3.6% mortality rate. Natural selection can generate favorable mutations with improved fitness advantages; however, the identified coronaviruses may be the tip of the iceberg, and potentially more fatal variants of concern (VOCs) may emerge over time. Understanding the patterns of emerging VOCs and forecasting mutations that may lead to gain of function or immune escape is urgently required. Here we developed PhyloTransformer, a Transformer-based discriminative model that engages a multi-head self-attention mechanism to model genetic mutations that may lead to viral reproductive advantage. In order to identify complex dependencies between the elements of each input sequence, PhyloTransformer utilizes advanced modeling techniques, including a novel Fast Attention Via positive Orthogonal Random features approach (FAVOR+) from Performer, and the Masked Language Model (MLM) from Bidirectional Encoder Representations from Transformers (BERT). PhyloTransformer was trained with 1,765,297 genetic sequences retrieved from the Global Initiative for Sharing All Influenza Data (GISAID) database. Firstly, we compared the prediction accuracy of novel mutations and novel combinations using extensive baseline models; we found that PhyloTransformer outperformed every baseline method with statistical significance. Secondly, we examined predictions of mutations in each nucleotide of the receptor binding motif (RBM), and we found our predictions were precise and accurate. Thirdly, we predicted modifications of N-glycosylation sites to identify mutations associated with altered glycosylation that may be favored during viral evolution. We anticipate that PhyloTransformer may guide proactive vaccine design for effective targeting of future SARS-CoV-2 variants.

Sequence Level Contrastive Learning for Text Summarization

Contrastive learning models have achieved great success in unsupervised visual representation learning, which maximize the similarities between feature representations of different views of the same image, while minimize the similarities between feature representations of views of different images. In text summarization, the output summary is a shorter form of the input document and they have similar meanings. In this paper, we propose a contrastive learning model for supervised abstractive text summarization, where we view a document, its gold summary and its model generated summaries as different views of the same mean representation and maximize the similarities between them during training. We improve over a strong sequence-to-sequence text generation model (i.e., BART) on three different summarization datasets. Human evaluation also shows that our model achieves better faithfulness ratings compared to its counterpart without contrastive objectives.

Unsupervised Extractive Summarization by Pre-training Hierarchical Transformers

Unsupervised extractive document summarization aims to select important sentences from a document without using labeled summaries during training. Existing methods are mostly graph-based with sentences as nodes and edge weights measured by sentence similarities. In this work, we find that transformer attentions can be used to rank sentences for unsupervised extractive summarization. Specifically, we first pre-train a hierarchical transformer model using unlabeled documents only. Then we propose a method to rank sentences using sentence-level self-attentions and pre-training objectives. Experiments on CNN/DailyMail and New York Times datasets show our model achieves state-of-the-art performance on unsupervised summarization. We also find in experiments that our model is less dependent on sentence positions. When using a linear combination of our model and a recent unsupervised model explicitly modeling sentence positions, we obtain even better results.