Tianle Li

Lianmin Zheng, Wei-Lin Chiang, Ying Sheng et al. · berkeley

Studying how people interact with large language models (LLMs) in real-world scenarios is increasingly important due to their widespread use in various applications. In this paper, we introduce LMSYS-Chat-1M, a large-scale dataset containing one million real-world conversations with 25 state-of-the-art LLMs. This dataset is collected from 210K unique IP addresses in the wild on our Vicuna demo and Chatbot Arena website. We offer an overview of the dataset's content, including its curation process, basic statistics, and topic distribution, highlighting its diversity, originality, and scale. We demonstrate its versatility through four use cases: developing content moderation models that perform similarly to GPT-4, building a safety benchmark, training instruction-following models that perform similarly to Vicuna, and creating challenging benchmark questions. We believe that this dataset will serve as a valuable resource for understanding and advancing LLM capabilities. The dataset is publicly available at https://huggingface.co/datasets/lmsys/lmsys-chat-1m.

Tianle Li, Max Ku, Cong Wei et al.

Subject-driven image generation aims at generating images containing customized subjects, which has recently drawn enormous attention from the research community. However, the previous works cannot precisely control the background and position of the target subject. In this work, we aspire to fill the void and propose two novel subject-driven sub-tasks, i.e., Subject Replacement and Subject Addition. The new tasks are challenging in multiple aspects: replacing a subject with a customized one can change its shape, texture, and color, while adding a target subject to a designated position in a provided scene necessitates a context-aware posture. To conquer these two novel tasks, we first manually curate a new dataset DreamEditBench containing 22 different types of subjects, and 440 source images with different difficulty levels. We plan to host DreamEditBench as a platform and hire trained evaluators for standard human evaluation. We also devise an innovative method DreamEditor to resolve these tasks by performing iterative generation, which enables a smooth adaptation to the customized subject. In this project, we conduct automatic and human evaluations to understand the performance of DreamEditor and baselines on DreamEditBench. For Subject Replacement, we found that the existing models are sensitive to the shape and color of the original subject. The model failure rate will dramatically increase when the source and target subjects are highly different. For Subject Addition, we found that the existing models cannot easily blend the customized subjects into the background smoothly, leading to noticeable artifacts in the generated image. We hope DreamEditBench can become a standard platform to enable future investigations toward building more controllable subject-driven image editing. Our project homepage is https://dreameditbenchteam.github.io/.

Tianle Li, Yushi Sun, Shang-ling Hsu et al.

The dissemination of fake news on social networks has drawn public need for effective and efficient fake news detection methods. Generally, fake news on social networks is multi-modal and has various connections with other entities such as users and posts. The heterogeneity in both news content and the relationship with other entities in social networks brings challenges to designing a model that comprehensively captures the local multi-modal semantics of entities in social networks and the global structural representation of the propagation patterns, so as to classify fake news effectively and accurately. In this paper, we propose a novel Transformer-based model: HetTransformer to solve the fake news detection problem on social networks, which utilises the encoder-decoder structure of Transformer to capture the structural information of news propagation patterns. We first capture the local heterogeneous semantics of news, post, and user entities in social networks. Then, we apply Transformer to capture the global structural representation of the propagation patterns in social networks for fake news detection. Experiments on three real-world datasets demonstrate that our model is able to outperform the state-of-the-art baselines in fake news detection.

Evan Frick, Tianle Li, Connor Chen et al. · berkeley

We introduce a new benchmark for reward models that quantifies their ability to produce strong language models through RLHF (Reinforcement Learning from Human Feedback). The gold-standard approach is to run a full RLHF training pipeline and directly probe downstream LLM performance. However, this process is prohibitively expensive. To address this, we build a predictive model of downstream LLM performance by evaluating the reward model on proxy tasks. These proxy tasks consist of a large-scale human preference and a verifiable correctness preference dataset, in which we measure 12 metrics across 12 domains. To investigate which reward model metrics are most correlated to gold-standard RLHF outcomes, we launch an end-to-end RLHF experiment on a large-scale crowdsourced human preference platform to view real reward model downstream performance as ground truth. Ultimately, we compile our data and findings into Preference Proxy Evaluations (PPE), the first reward model benchmark explicitly linked to post-RLHF real-world human preference performance, which we open-source for public use and further development. Our code and evaluations can be found at https://github.com/lmarena/PPE .

Evan Frick, Connor Chen, Joseph Tennyson et al.

Large language model (LLM) evaluations typically rely on aggregated metrics like accuracy or human preference, averaging across users and prompts. This averaging obscures user- and prompt-specific variations in model performance. To address this, we propose Prompt-to-Leaderboard (P2L), a method that produces leaderboards specific to a prompt. The core idea is to train an LLM taking natural language prompts as input to output a vector of Bradley-Terry coefficients which are then used to predict the human preference vote. The resulting prompt-dependent leaderboards allow for unsupervised task-specific evaluation, optimal routing of queries to models, personalization, and automated evaluation of model strengths and weaknesses. Data from Chatbot Arena suggest that P2L better captures the nuanced landscape of language model performance than the averaged leaderboard. Furthermore, our findings suggest that P2L's ability to produce prompt-specific evaluations follows a power law scaling similar to that observed in LLMs themselves. In January 2025, the router we trained based on this methodology achieved the #1 spot on the Chatbot Arena leaderboard. Our code is available on GitHub at https://github.com/lmarena/p2l.

Tianle Li, Ge Zhang, Quy Duc Do et al.

Large Language Models (LLMs) have made significant strides in handling long sequences. Some models like Gemini could even to be capable of dealing with millions of tokens. However, their performance evaluation has largely been confined to metrics like perplexity and synthetic tasks, which may not fully capture their true abilities in more challenging, real-world scenarios. We introduce a benchmark (LongICLBench) for long in-context learning in extreme-label classification using six datasets with 28 to 174 classes and input lengths from 2K to 50K tokens. Our benchmark requires LLMs to comprehend the entire input to recognize the massive label spaces to make correct predictions. We evaluate on 15 long-context LLMs and find that they perform well on less challenging classification tasks with smaller label space and shorter demonstrations. However, they struggle with more challenging task like Discovery with 174 labels, suggesting a gap in their ability to process long, context-rich sequences. Further analysis reveals a bias towards labels presented later in the sequence and a need for improved reasoning over multiple pieces of information. Our study reveals that long context understanding and reasoning is still a challenging task for the existing LLMs. We believe LongICLBench could serve as a more realistic evaluation for the future long-context LLMs.

Tianle Li, Yi Yang

NLP researchers propose different word-substitute black-box attacks that can fool text classification models. In such attack, an adversary keeps sending crafted adversarial queries to the target model until it can successfully achieve the intended outcome. State-of-the-art attack methods usually require hundreds or thousands of queries to find one adversarial example. In this paper, we study whether a sophisticated adversary can attack the system with much less queries. We propose a simple yet efficient method that can reduce the average number of adversarial queries by 3-30 times and maintain the attack effectiveness. This research highlights that an adversary can fool a deep NLP model with much less cost.

Dongfu Jiang, Max Ku, Tianle Li et al.

Generative AI has made remarkable strides to revolutionize fields such as image and video generation. These advancements are driven by innovative algorithms, architecture, and data. However, the rapid proliferation of generative models has highlighted a critical gap: the absence of trustworthy evaluation metrics. Current automatic assessments such as FID, CLIP, FVD, etc often fail to capture the nuanced quality and user satisfaction associated with generative outputs. This paper proposes an open platform GenAI-Arena to evaluate different image and video generative models, where users can actively participate in evaluating these models. By leveraging collective user feedback and votes, GenAI-Arena aims to provide a more democratic and accurate measure of model performance. It covers three tasks of text-to-image generation, text-to-video generation, and image editing respectively. Currently, we cover a total of 35 open-source generative models. GenAI-Arena has been operating for seven months, amassing over 9000 votes from the community. We describe our platform, analyze the data, and explain the statistical methods for ranking the models. To further promote the research in building model-based evaluation metrics, we release a cleaned version of our preference data for the three tasks, namely GenAI-Bench. We prompt the existing multi-modal models like Gemini, and GPT-4o to mimic human voting. We compute the accuracy by comparing the model voting with the human voting to understand their judging abilities. Our results show existing multimodal models are still lagging in assessing the generated visual content, even the best model GPT-4o only achieves an average accuracy of 49.19 across the three generative tasks. Open-source MLLMs perform even worse due to the lack of instruction-following and reasoning ability in complex vision scenarios.

Tianle Li, Xueguang Ma, Alex Zhuang et al.

Question answering over knowledge bases is considered a difficult problem due to the challenge of generalizing to a wide variety of possible natural language questions. Additionally, the heterogeneity of knowledge base schema items between different knowledge bases often necessitates specialized training for different knowledge base question-answering (KBQA) datasets. To handle questions over diverse KBQA datasets with a unified training-free framework, we propose KB-BINDER, which for the first time enables few-shot in-context learning over KBQA tasks. Firstly, KB-BINDER leverages large language models like Codex to generate logical forms as the draft for a specific question by imitating a few demonstrations. Secondly, KB-BINDER grounds on the knowledge base to bind the generated draft to an executable one with BM25 score matching. The experimental results on four public heterogeneous KBQA datasets show that KB-BINDER can achieve a strong performance with only a few in-context demonstrations. Especially on GraphQA and 3-hop MetaQA, KB-BINDER can even outperform the state-of-the-art trained models. On GrailQA and WebQSP, our model is also on par with other fully-trained models. We believe KB-BINDER can serve as an important baseline for future research. Our code is available at https://github.com/ltl3A87/KB-BINDER.

Xianchao Zeng, Xinyu Zhou, Youcheng Li et al.

Vision-Language-Action (VLA) models have recently achieved remarkable progress in robotic manipulation, yet they remain limited in failure diagnosis and learning from failures. Additionally, existing failure datasets are mostly generated programmatically in simulation, which limits their generalization to the real world. In light of these, we introduce ViFailback, a framework designed to diagnose robotic manipulation failures and provide both textual and visual correction guidance. Our framework utilizes explicit visual symbols to enhance annotation efficiency. We further release the ViFailback dataset, a large-scale collection of 58,126 Visual Question Answering (VQA) pairs along with their corresponding 5,202 real-world manipulation trajectories. Based on the dataset, we establish ViFailback-Bench, a benchmark of 11 fine-grained VQA tasks designed to assess the failure diagnosis and correction abilities of Vision-Language Models (VLMs), featuring ViFailback-Bench Lite for closed-ended and ViFailback-Bench Hard for open-ended evaluation. To demonstrate the effectiveness of our framework, we built the ViFailback-8B VLM, which not only achieves significant overall performance improvement on ViFailback-Bench but also generates visual symbols for corrective action guidance. Finally, by integrating ViFailback-8B with a VLA model, we conduct real-world robotic experiments demonstrating its ability to assist the VLA model in recovering from failures. Project Website: https://x1nyuzhou.github.io/vifailback.github.io/

Jihai Zhang, Tianle Li, Linjie Li et al.

Recent advancements in unified vision-language models (VLMs), which integrate both visual understanding and generation capabilities, have attracted significant attention. The underlying hypothesis is that a unified architecture with mixed training on both understanding and generation tasks can enable mutual enhancement between understanding and generation. However, this hypothesis remains underexplored in prior works on unified VLMs. To address this gap, this paper systematically investigates the generalization across understanding and generation tasks in unified VLMs. Specifically, we design a dataset closely aligned with real-world scenarios to facilitate extensive experiments and quantitative evaluations. We evaluate multiple unified VLM architectures to validate our findings. Our key findings are as follows. First, unified VLMs trained with mixed data exhibit mutual benefits in understanding and generation tasks across various architectures, and this mutual benefits can scale up with increased data. Second, better alignment between multimodal input and output spaces will lead to better generalization. Third, the knowledge acquired during generation tasks can transfer to understanding tasks, and this cross-task generalization occurs within the base language model, beyond modality adapters. Our findings underscore the critical necessity of unifying understanding and generation in VLMs, offering valuable insights for the design and optimization of unified VLMs.

Xingwu Chen, Tianle Li, Difan Zou

While reinforcement learning (RL) demonstrated remarkable success in enhancing the reasoning capabilities of language models, the training dynamics of RL in LLMs remain unclear. In this work, we provide an explanation of the RL training process through empirical analysis and rigorous theoretical modeling. First, through systematic reasoning-pattern-level and token-level analysis across the RL training process, we show that while different reasoning patterns exhibit relatively stable success rates during training, RL primarily optimizes a sparse subset of critical tokens, thereby reshaping reasoning pattern distributions to affect model performance. Building on these empirical insights, we develop a theoretical framework to understand the training dynamics of RL with two typical rewards: verifiable reward (RLVR) and model's internal feedback (RLIF). For RLVR, we analyze the training dynamics under two special cases: one where models readily converge to optimal reasoning strategies, and another where optimization becomes challenging, revealing that the base model's reasoning quality is crucial for determining convergence behavior. For RLIF, we examine how internal rewards initially improve model performance but can potentially lead to degradation with continued training. Extensive experiments validate our findings, advancing both theoretical understanding and practical applications of RL in language model enhancement.

Tianle Li, Jihai Zhang, Yongming Rao et al.

While large language models (LLMs) demonstrate strong reasoning capabilities utilizing reinforcement learning (RL) with verifiable reward, whether large vision-language models (VLMs) can directly inherit such capabilities through similar post-training strategies remains underexplored. In this work, we conduct a systematic compositional probing study to evaluate whether current VLMs trained with RL or other post-training strategies can compose capabilities across modalities or tasks under out-of-distribution conditions. We design a suite of diagnostic tasks that train models on unimodal tasks or isolated reasoning skills, and evaluate them on multimodal, compositional variants requiring skill integration. Through comparisons between supervised fine-tuning (SFT) and RL-trained models, we identify three key findings: (1) RL-trained models consistently outperform SFT on compositional generalization, demonstrating better integration of learned skills; (2) although VLMs achieve strong performance on individual tasks, they struggle to generalize compositionally under cross-modal and cross-task scenario, revealing a significant gap in current training strategies; (3) enforcing models to explicitly describe visual content before reasoning (e.g., caption-before-thinking), along with rewarding progressive vision-to-text grounding, yields notable gains. It highlights two essential ingredients for improving compositionality in VLMs: visual-to-text alignment and accurate visual grounding. Our findings shed light on the current limitations of RL-based reasoning VLM training and provide actionable insights toward building models that reason compositionally across modalities and tasks.

Lucas Spangher, Tianle Li, William F. Arnold et al.

There exists an extremely wide array of LLM benchmarking tasks, whereas oftentimes a single number is the most actionable for decision-making, especially by non-experts. No such aggregation schema exists that is not Elo-based, which could be costly or time-consuming. Here we propose a method to aggregate performance across a general space of benchmarks, nicknamed Project "MPG," dubbed Model Performance and Goodness, additionally referencing a metric widely understood to be an important yet inaccurate and crude measure of car performance. Here, we create two numbers: a "Goodness" number (answer accuracy) and a "Fastness" number (cost or QPS). We compare models against each other and present a ranking according to our general metric as well as subdomains. We find significant agreement between the raw Pearson correlation of our scores and those of Chatbot Arena, even improving on the correlation of the MMLU leaderboard to Chatbot Arena.

Tianle Li, Wei-Lin Chiang, Evan Frick et al.

The rapid evolution of Large Language Models (LLMs) has outpaced the development of model evaluation, highlighting the need for continuous curation of new, challenging benchmarks. However, manual curation of high-quality, human-aligned benchmarks is expensive and time-consuming. To address this, we introduce BenchBuilder, an automated pipeline that leverages LLMs to curate high-quality, open-ended prompts from large, crowd-sourced datasets, enabling continuous benchmark updates without human in the loop. We apply BenchBuilder to datasets such as Chatbot Arena and WildChat-1M, extracting challenging prompts and utilizing LLM-as-a-Judge for automatic model evaluation. To validate benchmark quality, we propose new metrics to measure a benchmark's alignment with human preferences and ability to separate models. We release Arena-Hard-Auto, a benchmark consisting 500 challenging prompts curated by BenchBuilder. Arena-Hard-Auto provides 3x higher separation of model performances compared to MT-Bench and achieves 98.6% correlation with human preference rankings, all at a cost of $20. Our work sets a new framework for the scalable curation of automated benchmarks from extensive data.

Yubo Wang, Xueguang Ma, Ge Zhang et al.

In the age of large-scale language models, benchmarks like the Massive Multitask Language Understanding (MMLU) have been pivotal in pushing the boundaries of what AI can achieve in language comprehension and reasoning across diverse domains. However, as models continue to improve, their performance on these benchmarks has begun to plateau, making it increasingly difficult to discern differences in model capabilities. This paper introduces MMLU-Pro, an enhanced dataset designed to extend the mostly knowledge-driven MMLU benchmark by integrating more challenging, reasoning-focused questions and expanding the choice set from four to ten options. Additionally, MMLU-Pro eliminates the trivial and noisy questions in MMLU. Our experimental results show that MMLU-Pro not only raises the challenge, causing a significant drop in accuracy by 16% to 33% compared to MMLU but also demonstrates greater stability under varying prompts. With 24 different prompt styles tested, the sensitivity of model scores to prompt variations decreased from 4-5% in MMLU to just 2% in MMLU-Pro. Additionally, we found that models utilizing Chain of Thought (CoT) reasoning achieved better performance on MMLU-Pro compared to direct answering, which is in stark contrast to the findings on the original MMLU, indicating that MMLU-Pro includes more complex reasoning questions. Our assessments confirm that MMLU-Pro is a more discriminative benchmark to better track progress in the field.