Weijia Xu

Weijia Xu, Sweta Agrawal, Eleftheria Briakou et al.

Neural sequence generation models are known to "hallucinate", by producing outputs that are unrelated to the source text. These hallucinations are potentially harmful, yet it remains unclear in what conditions they arise and how to mitigate their impact. In this work, we first identify internal model symptoms of hallucinations by analyzing the relative token contributions to the generation in contrastive hallucinated vs. non-hallucinated outputs generated via source perturbations. We then show that these symptoms are reliable indicators of natural hallucinations, by using them to design a lightweight hallucination detector which outperforms both model-free baselines and strong classifiers based on quality estimation or large pre-trained models on manually annotated English-Chinese and German-English translation test beds.

Sudha Rao, Weijia Xu, Michael Xu et al.

The use of generative AI in video game development is on the rise, and as the conversational and other capabilities of large language models continue to improve, we expect LLM-driven non-player characters (NPCs) to become widely deployed. In this paper, we seek to understand how human players collaborate with LLM-driven NPCs to accomplish in-game goals. We design a minigame within Minecraft where a player works with two GPT4-driven NPCs to complete a quest. We perform a user study in which 28 Minecraft players play this minigame and share their feedback. On analyzing the game logs and recordings, we find that several patterns of collaborative behavior emerge from the NPCs and the human players. We also report on the current limitations of language-only models that do not have rich game-state or visual understanding. We believe that this preliminary study and analysis will inform future game developers on how to better exploit these rapidly improving generative AI models for collaborative roles in games.

Jorge Leandro, Sudha Rao, Michael Xu et al.

Dialogue-based Role Playing Games (RPGs) require powerful storytelling. The narratives of these may take years to write and typically involve a large creative team. In this work, we demonstrate the potential of large generative text models to assist this process. \textbf{GENEVA}, a prototype tool, generates a rich narrative graph with branching and reconverging storylines that match a high-level narrative description and constraints provided by the designer. A large language model (LLM), GPT-4, is used to generate the branching narrative and to render it in a graph format in a two-step process. We illustrate the use of GENEVA in generating new branching narratives for four well-known stories under different contextual constraints. This tool has the potential to assist in game development, simulations, and other applications with game-like properties.

Wanru Zhao, Lucas Caccia, Zhengyan Shi et al.

Curriculum learning is a class of training strategies that organizes the data being exposed to a model by difficulty, gradually from simpler to more complex examples. This research explores a reverse curriculum generation approach that recursively decomposes complex datasets into simpler, more learnable components. We propose a teacher-student framework where the teacher is equipped with the ability to reason step-by-step, which is used to recursively generate easier versions of examples, enabling the student model to progressively master difficult tasks. We propose a novel scoring system to measure data difficulty based on its structural complexity and conceptual depth, allowing curriculum construction over decomposed data. Experiments on math datasets (MATH and AIME) and code generation datasets demonstrate that models trained with curricula generated by our approach exhibit superior performance compared to standard training on original datasets.

An Yang, Bowen Yu, Chengyuan Li et al.

We introduce Qwen2.5-1M, a series of models that extend the context length to 1 million tokens. Compared to the previous 128K version, the Qwen2.5-1M series have significantly enhanced long-context capabilities through long-context pre-training and post-training. Key techniques such as long data synthesis, progressive pre-training, and multi-stage supervised fine-tuning are employed to effectively enhance long-context performance while reducing training costs. To promote the use of long-context models among a broader user base, we present and open-source our inference framework. This framework includes a length extrapolation method that can expand the model context lengths by at least four times, or even more, without additional training. To reduce inference costs, we implement a sparse attention method along with chunked prefill optimization for deployment scenarios and a sparsity refinement method to improve precision. Additionally, we detail our optimizations in the inference engine, including kernel optimization, pipeline parallelism, and scheduling optimization, which significantly enhance overall inference performance. By leveraging our inference framework, the Qwen2.5-1M models achieve a remarkable 3x to 7x prefill speedup in scenarios with 1 million tokens of context. This framework provides an efficient and powerful solution for developing applications that require long-context processing using open-source models. The Qwen2.5-1M series currently includes the open-source models Qwen2.5-7B-Instruct-1M and Qwen2.5-14B-Instruct-1M, as well as the API-accessed model Qwen2.5-Turbo. Evaluations show that Qwen2.5-1M models have been greatly improved in long-context tasks without compromising performance in short-context scenarios. Specifically, the Qwen2.5-14B-Instruct-1M model significantly outperforms GPT-4o-mini in long-context tasks and supports contexts eight times longer.

Weijia Xu, Alessandro Sordoni, Chandan Singh et al.

We introduce EvoLib, a test-time learning framework that enables large language models to accumulate, reuse, and evolve knowledge across problem instances without parameter updates or external supervision. Instead of adapting model parameters, our approach maintains a shared library of knowledge abstractions, including modular skills and reflective insights, automatically extracted from the model's own inference trajectories. To support continual improvement, we introduce a principled weighting and consolidation mechanism that jointly optimizes for immediate utility and long-term value. This allows simple, instance-specific abstractions to evolve into more general and reusable ones over time. Across challenging benchmarks in mathematical reasoning, code generation, and multi-turn agentic environments, EvoLib improves substantially over the top test-time scaling and learning methods without ground-truth feedback.

Zihan Jiang, Wanling Gao, Lei Wang et al.

In recent years, with the trend of applying deep learning (DL) in high performance scientific computing, the unique characteristics of emerging DL workloads in HPC raise great challenges in designing, implementing HPC AI systems. The community needs a new yard stick for evaluating the future HPC systems. In this paper, we propose HPC AI500 --- a benchmark suite for evaluating HPC systems that running scientific DL workloads. Covering the most representative scientific fields, each workload from HPC AI500 is based on real-world scientific DL applications. Currently, we choose 14 scientific DL benchmarks from perspectives of application scenarios, data sets, and software stack. We propose a set of metrics for comprehensively evaluating the HPC AI systems, considering both accuracy, performance as well as power and cost. We provide a scalable reference implementation of HPC AI500. HPC AI500 is a part of the open-source AIBench project, the specification and source code are publicly available from \url{http://www.benchcouncil.org/AIBench/index.html}.

Chandan Singh, Yan Shuo Tan, Weijia Xu et al.

Agentic data science (ADS) systems are rapidly improving their capability to autonomously analyze, fit, and interpret data, potentially moving towards a future where agents conduct the vast majority of data-science work. However, current ADS systems use statistical tools designed to be interpretable by humans, rather than interpretable by agents. To address this, we introduce Agentic-imodels, an agentic autoresearch loop that evolves data-science tools designed to be interpretable by agents. Specifically, it develops a library of scikit-learn-compatible regressors for tabular data that are optimized for both predictive performance and a novel LLM-based interpretability metric. The metric measures a suite of LLM-graded tests that probe whether a fitted model's string representation is "simulatable" by an LLM, i.e. whether the LLM can answer questions about the model's behavior by reading its string output alone. We find that the evolved models jointly improve predictive performance and agent-facing interpretability, generalizing to new datasets and new interpretability tests. Furthermore, these evolved models improve downstream end-to-end ADS, increasing performance for Copilot CLI, Claude Code, and Codex on the BLADE benchmark by up to 73%

Xiangyu Peng, Jessica Quaye, Sudha Rao et al.

We explore how interaction with large language models (LLMs) can give rise to emergent behaviors, empowering players to participate in the evolution of game narratives. Our testbed is a text-adventure game in which players attempt to solve a mystery under a fixed narrative premise, but can freely interact with non-player characters generated by GPT-4, a large language model. We recruit 28 gamers to play the game and use GPT-4 to automatically convert the game logs into a node-graph representing the narrative in the player's gameplay. We find that through their interactions with the non-deterministic behavior of the LLM, players are able to discover interesting new emergent nodes that were not a part of the original narrative but have potential for being fun and engaging. Players that created the most emergent nodes tended to be those that often enjoy games that facilitate discovery, exploration and experimentation.

Weijia Xu, Nebojsa Jojic, Sudha Rao et al.

With rapid advances in large language models (LLMs), there has been an increasing application of LLMs in creative content ideation and generation. A critical question emerges: can current LLMs provide ideas that are diverse enough to truly bolster collective creativity? We examine two state-of-the-art LLMs, GPT-4 and LLaMA-3, on story generation and discover that LLM-generated stories often consist of plot elements that are echoed across a number of generations. To quantify this phenomenon, we introduce the Sui Generis score, an automatic metric that measures the uniqueness of a plot element among alternative storylines generated using the same prompt under an LLM. Evaluating on 100 short stories, we find that LLM-generated stories often contain combinations of idiosyncratic plot elements echoed frequently across generations and across different LLMs, while plots from the original human-written stories are rarely recreated or even echoed in pieces. Moreover, our human evaluation shows that the ranking of Sui Generis scores among story segments correlates moderately with human judgment of surprise level, even though score computation is completely automatic without relying on human judgment.

Seyed Hossein Alavi, Weijia Xu, Nebojsa Jojic et al.

We introduce GamePlot, an LLM-powered assistant that supports game designers in crafting immersive narratives for turn-based games, and allows them to test these games through a collaborative game play and refine the plot throughout the process. Our user study with 14 game designers shows high levels of both satisfaction with the generated game plots and sense of ownership over the narratives, but also reconfirms that LLM are limited in their ability to generate complex and truly innovative content. We also show that diverse user populations have different expectations from AI assistants, and encourage researchers to study how tailoring assistants to diverse user groups could potentially lead to increased job satisfaction and greater creativity and innovation over time.

Weijia Xu, Andrzej Banburski-Fahey, Nebojsa Jojic

We introduce Reprompting, an iterative sampling algorithm that automatically learns the Chain-of-Thought (CoT) recipes for a given task without human intervention. Through Gibbs sampling, Reprompting infers the CoT recipes that work consistently well for a set of training samples by iteratively sampling new recipes using previously sampled recipes as parent prompts to solve other training problems. We conduct extensive experiments on 20 challenging reasoning tasks. Results show that Reprompting outperforms human-written CoT prompts substantially by +9.4 points on average. It also achieves consistently better performance than the state-of-the-art prompt optimization and decoding algorithms.

Shaolun Ruan, Yong Wang, Hailong Jiang et al.

Cloud systems are becoming increasingly powerful and complex. It is highly challenging to identify anomalous execution behaviors and pinpoint problems by examining the overwhelming intermediate results/states in complex application workflows. Domain scientists urgently need a friendly and functional interface to understand the quality of the computing services and the performance of their applications in real time. To meet these needs, we explore data generated by job schedulers and investigate general performance metrics (e.g., utilization of CPU, memory and disk I/O). Specifically, we propose an interactive visual analytics approach, BatchLens, to provide both providers and users of cloud service with an intuitive and effective way to explore the status of system batch jobs and help them conduct root-cause analysis of anomalous behaviors in batch jobs. We demonstrate the effectiveness of BatchLens through a case study on the public Alibaba bench workload trace datasets.

Weijia Xu, Heidi Ross, Joel Meyer et al.

Pedestrian safety is a priority for transportation system managers and operators, and a main focus of the Vision Zero strategy employed by the City of Austin, Texas. While there are a number of treatments and technologies to effectively improve pedestrian safety, identifying the location where these treatments are most needed remains a challenge. Current practice requires manual observation of candidate locations for limited time periods, leading to an identification process that is time consuming, lags behind traffic pattern changes over time, and lacks scalability. Mid-block locations, where safety countermeasures are often needed the most, are especially hard to identify and monitor. The goal for this research is to understand the correlation between bus stop locations and mid-block crossings, so as to assist traffic engineers in implementing Vision Zero strategies to improve pedestrian safety. In a prior work, we have developed a tool to detect pedestrian crossing events with traffic camera video using a deep neural network model to identify crossing events. In this paper, we extend the methods to identify bus stop usage with traffic camera video from off-the-shelf CCTV pan-tilt-zoom (PTZ) traffic monitoring cameras installed at nearby intersections. We correlate the video detection results for mid-block crossings near a bus stop, with pedestrian activity at the bus stops in each side of the mid-block crossing. We also implement a web portal to facilitate manual review of pedestrian activity detections by automating creation of video clips that show only crossing events, thereby vastly improving the efficiency of the human review process.

Weijia Xu, Marine Carpuat

Current approaches to incorporating terminology constraints in machine translation (MT) typically assume that the constraint terms are provided in their correct morphological forms. This limits their application to real-world scenarios where constraint terms are provided as lemmas. In this paper, we introduce a modular framework for incorporating lemma constraints in neural MT (NMT) in which linguistic knowledge and diverse types of NMT models can be flexibly applied. It is based on a novel cross-lingual inflection module that inflects the target lemma constraints based on the source context. We explore linguistically motivated rule-based and data-driven neural-based inflection modules and design English-German health and English-Lithuanian news test suites to evaluate them in domain adaptation and low-resource MT settings. Results show that our rule-based inflection module helps NMT models incorporate lemma constraints more accurately than a neural module and outperforms the existing end-to-end approach with lower training costs.

Weijia Xu, Batool Haider, Jason Krone et al.

Multilingual pre-trained contextual embedding models (Devlin et al., 2019) have achieved impressive performance on zero-shot cross-lingual transfer tasks. Finding the most effective fine-tuning strategy to fine-tune these models on high-resource languages so that it transfers well to the zero-shot languages is a non-trivial task. In this paper, we propose a novel meta-optimizer to soft-select which layers of the pre-trained model to freeze during fine-tuning. We train the meta-optimizer by simulating the zero-shot transfer scenario. Results on cross-lingual natural language inference show that our approach improves over the simple fine-tuning baseline and X-MAML (Nooralahzadeh et al., 2020).

Weijia Xu, Shuming Ma, Dongdong Zhang et al.

While non-autoregressive (NAR) models are showing great promise for machine translation, their use is limited by their dependence on knowledge distillation from autoregressive models. To address this issue, we seek to understand why distillation is so effective. Prior work suggests that distilled training data is less complex than manual translations. Based on experiments with the Levenshtein Transformer and the Mask-Predict NAR models on the WMT14 German-English task, this paper shows that different types of complexity have different impacts: while reducing lexical diversity and decreasing reordering complexity both help NAR learn better alignment between source and target, and thus improve translation quality, lexical diversity is the main reason why distillation increases model confidence, which affects the calibration of different NAR models differently.

Weijia Xu, Marine Carpuat

We introduce an Edit-Based Transformer with Repositioning (EDITOR), which makes sequence generation flexible by seamlessly allowing users to specify preferences in output lexical choice. Building on recent models for non-autoregressive sequence generation (Gu et al., 2019), EDITOR generates new sequences by iteratively editing hypotheses. It relies on a novel reposition operation designed to disentangle lexical choice from word positioning decisions, while enabling efficient oracles for imitation learning and parallel edits at decoding time. Empirically, EDITOR uses soft lexical constraints more effectively than the Levenshtein Transformer (Gu et al., 2019) while speeding up decoding dramatically compared to constrained beam search (Post and Vilar, 2018). EDITOR also achieves comparable or better translation quality with faster decoding speed than the Levenshtein Transformer on standard Romanian-English, English-German, and English-Japanese machine translation tasks.

Weijia Xu, Xing Niu, Marine Carpuat

While Iterative Back-Translation and Dual Learning effectively incorporate monolingual training data in neural machine translation, they use different objectives and heuristic gradient approximation strategies, and have not been extensively compared. We introduce a novel dual reconstruction objective that provides a unified view of Iterative Back-Translation and Dual Learning. It motivates a theoretical analysis and controlled empirical study on German-English and Turkish-English tasks, which both suggest that Iterative Back-Translation is more effective than Dual Learning despite its relative simplicity.

J. Gregory Pauloski, Zhao Zhang, Lei Huang et al.

Training neural networks with many processors can reduce time-to-solution; however, it is challenging to maintain convergence and efficiency at large scales. The Kronecker-factored Approximate Curvature (K-FAC) was recently proposed as an approximation of the Fisher Information Matrix that can be used in natural gradient optimizers. We investigate here a scalable K-FAC design and its applicability in convolutional neural network (CNN) training at scale. We study optimization techniques such as layer-wise distribution strategies, inverse-free second-order gradient evaluation, and dynamic K-FAC update decoupling to reduce training time while preserving convergence. We use residual neural networks (ResNet) applied to the CIFAR-10 and ImageNet-1k datasets to evaluate the correctness and scalability of our K-FAC gradient preconditioner. With ResNet-50 on the ImageNet-1k dataset, our distributed K-FAC implementation converges to the 75.9% MLPerf baseline in 18-25% less time than does the classic stochastic gradient descent (SGD) optimizer across scales on a GPU cluster.

Weijia Xu, Batool Haider, Saab Mansour

Natural language understanding (NLU) in the context of goal-oriented dialog systems typically includes intent classification and slot labeling tasks. Existing methods to expand an NLU system to new languages use machine translation with slot label projection from source to the translated utterances, and thus are sensitive to projection errors. In this work, we propose a novel end-to-end model that learns to align and predict target slot labels jointly for cross-lingual transfer. We introduce MultiATIS++, a new multilingual NLU corpus that extends the Multilingual ATIS corpus to nine languages across four language families, and evaluate our method using the corpus. Results show that our method outperforms a simple label projection method using fast-align on most languages, and achieves competitive performance to the more complex, state-of-the-art projection method with only half of the training time. We release our MultiATIS++ corpus to the community to continue future research on cross-lingual NLU.

Weijia Xu, Xing Niu, Marine Carpuat

Despite some empirical success at correcting exposure bias in machine translation, scheduled sampling algorithms suffer from a major drawback: they incorrectly assume that words in the reference translations and in sampled sequences are aligned at each time step. Our new differentiable sampling algorithm addresses this issue by optimizing the probability that the reference can be aligned with the sampled output, based on a soft alignment predicted by the model itself. As a result, the output distribution at each time step is evaluated with respect to the whole predicted sequence. Experiments on IWSLT translation tasks show that our approach improves BLEU compared to maximum likelihood and scheduled sampling baselines. In addition, our approach is simpler to train with no need for sampling schedule and yields models that achieve larger improvements with smaller beam sizes.

Xing Niu, Weijia Xu, Marine Carpuat

We aim to better exploit the limited amounts of parallel text available in low-resource settings by introducing a differentiable reconstruction loss for neural machine translation (NMT). This loss compares original inputs to reconstructed inputs, obtained by back-translating translation hypotheses into the input language. We leverage differentiable sampling and bi-directional NMT to train models end-to-end, without introducing additional parameters. This approach achieves small but consistent BLEU improvements on four language pairs in both translation directions, and outperforms an alternative differentiable reconstruction strategy based on hidden states.