Tianqi Liu

Zhen Qin, Rolf Jagerman, Kai Hui et al. · deepmind

Ranking documents using Large Language Models (LLMs) by directly feeding the query and candidate documents into the prompt is an interesting and practical problem. However, researchers have found it difficult to outperform fine-tuned baseline rankers on benchmark datasets. We analyze pointwise and listwise ranking prompts used by existing methods and argue that off-the-shelf LLMs do not fully understand these challenging ranking formulations. In this paper, we propose to significantly reduce the burden on LLMs by using a new technique called Pairwise Ranking Prompting (PRP). Our results are the first in the literature to achieve state-of-the-art ranking performance on standard benchmarks using moderate-sized open-sourced LLMs. On TREC-DL 2019&2020, PRP based on the Flan-UL2 model with 20B parameters performs favorably with the previous best approach in the literature, which is based on the blackbox commercial GPT-4 that has 50x (estimated) model size, while outperforming other LLM-based solutions, such as InstructGPT which has 175B parameters, by over 10% for all ranking metrics. By using the same prompt template on seven BEIR tasks, PRP outperforms supervised baselines and outperforms the blackbox commercial ChatGPT solution by 4.2% and pointwise LLM-based solutions by more than 10% on average NDCG@10. Furthermore, we propose several variants of PRP to improve efficiency and show that it is possible to achieve competitive results even with linear complexity.

Tianqi Liu, Xinyi Ye, Weiyue Zhao et al.

Learning-based multi-view stereo (MVS) method heavily relies on feature matching, which requires distinctive and descriptive representations. An effective solution is to apply non-local feature aggregation, e.g., Transformer. Albeit useful, these techniques introduce heavy computation overheads for MVS. Each pixel densely attends to the whole image. In contrast, we propose to constrain non-local feature augmentation within a pair of lines: each point only attends the corresponding pair of epipolar lines. Our idea takes inspiration from the classic epipolar geometry, which shows that one point with different depth hypotheses will be projected to the epipolar line on the other view. This constraint reduces the 2D search space into the epipolar line in stereo matching. Similarly, this suggests that the matching of MVS is to distinguish a series of points lying on the same line. Inspired by this point-to-line search, we devise a line-to-point non-local augmentation strategy. We first devise an optimized searching algorithm to split the 2D feature maps into epipolar line pairs. Then, an Epipolar Transformer (ET) performs non-local feature augmentation among epipolar line pairs. We incorporate the ET into a learning-based MVS baseline, named ET-MVSNet. ET-MVSNet achieves state-of-the-art reconstruction performance on both the DTU and Tanks-and-Temples benchmark with high efficiency. Code is available at https://github.com/TQTQliu/ET-MVSNet.

Wei Xiong, Chengshuai Shi, Jiaming Shen et al.

Recent studies have shown that large language models' (LLMs) mathematical problem-solving capabilities can be enhanced by integrating external tools, such as code interpreters, and employing multi-turn Chain-of-Thought (CoT) reasoning. While current methods focus on synthetic data generation and Supervised Fine-Tuning (SFT), this paper studies the complementary direct preference learning approach to further improve model performance. However, existing direct preference learning algorithms are originally designed for the single-turn chat task, and do not fully address the complexities of multi-turn reasoning and external tool integration required for tool-integrated mathematical reasoning tasks. To fill in this gap, we introduce a multi-turn direct preference learning framework, tailored for this context, that leverages feedback from code interpreters and optimizes trajectory-level preferences. This framework includes multi-turn DPO and multi-turn KTO as specific implementations. The effectiveness of our framework is validated through training of various language models using an augmented prompt set from the GSM8K and MATH datasets. Our results demonstrate substantial improvements: a supervised fine-tuned Gemma-1.1-it-7B model's performance increased from 77.5% to 83.9% on GSM8K and from 46.1% to 51.2% on MATH. Similarly, a Gemma-2-it-9B model improved from 84.1% to 86.3% on GSM8K and from 51.0% to 54.5% on MATH.

Tianqi Liu, Yao Zhao, Rishabh Joshi et al.

Improving the alignment of language models with human preferences remains an active research challenge. Previous approaches have primarily utilized Reinforcement Learning from Human Feedback (RLHF) via online RL methods such as Proximal Policy Optimization (PPO). Recently, offline methods such as Sequence Likelihood Calibration (SLiC) and Direct Preference Optimization (DPO) have emerged as attractive alternatives, offering improvements in stability and scalability while maintaining competitive performance. SLiC refines its loss function using sequence pairs sampled from a supervised fine-tuned (SFT) policy, while DPO directly optimizes language models based on preference data, foregoing the need for a separate reward model. However, the maximum likelihood estimator (MLE) of the target optimal policy requires labeled preference pairs sampled from that policy. DPO's lack of a reward model constrains its ability to sample preference pairs from the optimal policy, and SLiC is restricted to sampling preference pairs only from the SFT policy. To address these limitations, we introduce a novel approach called Statistical Rejection Sampling Optimization (RSO) that aims to source preference data from the target optimal policy using rejection sampling, enabling a more accurate estimation of the optimal policy. We also propose a unified framework that enhances the loss functions used in both SLiC and DPO from a preference modeling standpoint. Through extensive experiments across three diverse tasks, we demonstrate that RSO consistently outperforms both SLiC and DPO on evaluations from both Large Language Model (LLM) and human raters.

Yue Yu, Jiaming Shen, Tianqi Liu et al.

Large language models (LLMs) have shown remarkable capabilities in various natural language understanding tasks. With only a few demonstration examples, these LLMs can quickly adapt to target tasks without expensive gradient updates. Common strategies to boost such 'in-context' learning ability are to ensemble multiple model decoded results and require the model to generate an explanation along with the prediction. However, these models often treat different class predictions equally and neglect the potential discrepancy between the explanations and predictions. To fully unleash the power of explanations, we propose EASE, an Explanation-Aware Soft Ensemble framework to empower in-context learning with LLMs. We design two techniques, explanation-guided ensemble, and soft probability aggregation, to mitigate the effect of unreliable explanations and improve the consistency between explanations and final predictions. Experiments on seven natural language understanding tasks and four varying-size LLMs demonstrate the effectiveness of our proposed framework.

Jiaming Shen, Tianqi Liu, Jialu Liu et al.

The popularity of automated news headline generation has surged with advancements in pre-trained language models. However, these models often suffer from the ``hallucination'' problem, where the generated headline is not fully supported by its source article. Efforts to address this issue have predominantly focused on English, using over-simplistic classification schemes that overlook nuanced hallucination types. In this study, we introduce the first multilingual, fine-grained news headline hallucination detection dataset that contains over 11 thousand pairs in 5 languages, each annotated with detailed hallucination types by experts. We conduct extensive experiments on this dataset under two settings. First, we implement several supervised fine-tuning approaches as preparatory solutions and demonstrate this dataset's challenges and utilities. Second, we test various large language models' in-context learning abilities and propose two novel techniques, language-dependent demonstration selection and coarse-to-fine prompting, to boost the few-shot hallucination detection performance in terms of the example-F1 metric. We release this dataset to foster further research in multilingual, fine-grained headline hallucination detection.

Jing Nathan Yan, Tianqi Liu, Justin T Chiu et al.

Comparative reasoning plays a crucial role in text preference prediction; however, large language models (LLMs) often demonstrate inconsistencies in their reasoning. While approaches like Chain-of-Thought improve accuracy in many other settings, they struggle to consistently distinguish the similarities and differences of complex texts. We introduce SC, a prompting approach that predicts text preferences by generating structured intermediate comparisons. SC begins by proposing aspects of comparison, followed by generating textual comparisons under each aspect. We select consistent comparisons with a pairwise consistency comparator that ensures each aspect's comparisons clearly distinguish differences between texts, significantly reducing hallucination and improving consistency. Our comprehensive evaluations across various NLP tasks, including summarization, retrieval, and automatic rating, demonstrate that SC equips LLMs to achieve state-of-the-art performance in text preference prediction.

Jiaxin Huang, Tianqi Liu, Jialu Liu et al.

Multi-Task Learning (MTL) models have shown their robustness, effectiveness, and efficiency for transferring learned knowledge across tasks. In real industrial applications such as web content classification, multiple classification tasks are predicted from the same input text such as a web article. However, at the serving time, the existing multitask transformer models such as prompt or adaptor based approaches need to conduct N forward passes for N tasks with O(N) computation cost. To tackle this problem, we propose a scalable method that can achieve stronger performance with close to O(1) computation cost via only one forward pass. To illustrate real application usage, we release a multitask dataset on news topic and style classification. Our experiments show that our proposed method outperforms strong baselines on both the GLUE benchmark and our news dataset. Our code and dataset are publicly available at https://bit.ly/mtop-code.

Zhen Qin, Junru Wu, Jiaming Shen et al.

We introduce LAMPO, a novel paradigm that leverages Large Language Models (LLMs) for solving few-shot multi-class ordinal classification tasks. Unlike conventional methods, which concatenate all demonstration examples with the test instance and prompt LLMs to produce the pointwise prediction, our framework uses the LLM as a preference machine that makes a relative comparative decision between the test instance and each demonstration. A self-supervised method is then introduced to aggregate these binary comparisons into the final ordinal decision. LAMPO addresses several limitations inherent in previous methods, including context length constraints, ordering biases, and challenges associated with absolute point-wise estimation. Extensive experiments on seven public datasets demonstrate LAMPO's remarkably competitive performance across a diverse spectrum of applications (e.g., movie review analysis and hate speech detection). Notably, in certain applications, the improvement can be substantial, exceeding 20% in an absolute term. Moreover, we believe LAMPO represents an interesting addition to the non-parametric application layered on top of LLMs, as it supports black-box LLMs without necessitating the outputting of LLM's internal states (e.g., embeddings), as seen in previous approaches.

Shangmin Guo, Biao Zhang, Tianlin Liu et al.

Direct alignment from preferences (DAP) methods, such as DPO, have recently emerged as efficient alternatives to reinforcement learning from human feedback (RLHF), that do not require a separate reward model. However, the preference datasets used in DAP methods are usually collected ahead of training and never updated, thus the feedback is purely offline. Moreover, responses in these datasets are often sampled from a language model distinct from the one being aligned, and since the model evolves over training, the alignment phase is inevitably off-policy. In this study, we posit that online feedback is key and improves DAP methods. Our method, online AI feedback (OAIF), uses an LLM as annotator: on each training iteration, we sample two responses from the current model and prompt the LLM annotator to choose which one is preferred, thus providing online feedback. Despite its simplicity, we demonstrate via human evaluation in several tasks that OAIF outperforms both offline DAP and RLHF methods. We further show that the feedback leveraged in OAIF is easily controllable, via instruction prompts to the LLM annotator.

Tianqi Liu, Bo Liu

This paper presents our 7th place solution to the second YouTube-8M video understanding competition which challenges participates to build a constrained-size model to classify millions of YouTube videos into thousands of classes. Our final model consists of four single models aggregated into one tensorflow graph. For each single model, we use the same network architecture as in the winning solution of the first YouTube-8M video understanding competition, namely Gated NetVLAD. We train the single models separately in tensorflow's default float32 precision, then replace weights with float16 precision and ensemble them in the evaluation and inference stages., achieving 48.5% compression rate without loss of precision. Our best model achieved 88.324% GAP on private leaderboard. The code is publicly available at https://github.com/boliu61/youtube-8m

Tianqi Liu, Zhen Qin, Junru Wu et al.

Aligning language models (LMs) with curated human feedback is critical to control their behaviors in real-world applications. Several recent policy optimization methods, such as DPO and SLiC, serve as promising alternatives to the traditional Reinforcement Learning from Human Feedback (RLHF) approach. In practice, human feedback often comes in a format of a ranked list over multiple responses to amortize the cost of reading prompt. Multiple responses can also be ranked by reward models or AI feedback. There lacks such a thorough study on directly fitting upon a list of responses. In this work, we formulate the LM alignment as a \textit{listwise} ranking problem and describe the LiPO framework, where the policy can potentially learn more effectively from a ranked list of plausible responses given the prompt. This view draws an explicit connection to Learning-to-Rank (LTR), where most existing preference optimization work can be mapped to existing ranking objectives. Following this connection, we provide an examination of ranking objectives that are not well studied for LM alignment with DPO and SLiC as special cases when list size is two. In particular, we highlight a specific method, LiPO-$λ$, which leverages a state-of-the-art \textit{listwise} ranking objective and weights each preference pair in a more advanced manner. We show that LiPO-$λ$ can outperform DPO variants and SLiC by a clear margin on several preference alignment tasks with both curated and real rankwise preference data.

Daniele Calandriello, Daniel Guo, Remi Munos et al.

Ensuring alignment of language models' outputs with human preferences is critical to guarantee a useful, safe, and pleasant user experience. Thus, human alignment has been extensively studied recently and several methods such as Reinforcement Learning from Human Feedback (RLHF), Direct Policy Optimisation (DPO) and Sequence Likelihood Calibration (SLiC) have emerged. In this paper, our contribution is two-fold. First, we show the equivalence between two recent alignment methods, namely Identity Policy Optimisation (IPO) and Nash Mirror Descent (Nash-MD). Second, we introduce a generalisation of IPO, named IPO-MD, that leverages the regularised sampling approach proposed by Nash-MD. This equivalence may seem surprising at first sight, since IPO is an offline method whereas Nash-MD is an online method using a preference model. However, this equivalence can be proven when we consider the online version of IPO, that is when both generations are sampled by the online policy and annotated by a trained preference model. Optimising the IPO loss with such a stream of data becomes then equivalent to finding the Nash equilibrium of the preference model through self-play. Building on this equivalence, we introduce the IPO-MD algorithm that generates data with a mixture policy (between the online and reference policy) similarly as the general Nash-MD algorithm. We compare online-IPO and IPO-MD to different online versions of existing losses on preference data such as DPO and SLiC on a summarisation task.

Hanguang Xiao, Feizhong Zhou, Xingyue Liu et al.

Since the release of ChatGPT and GPT-4, large language models (LLMs) and multimodal large language models (MLLMs) have attracted widespread attention for their exceptional capabilities in understanding, reasoning, and generation, introducing transformative paradigms for integrating artificial intelligence into medicine. This survey provides a comprehensive overview of the development, principles, application scenarios, challenges, and future directions of LLMs and MLLMs in medicine. Specifically, it begins by examining the paradigm shift, tracing the transition from traditional models to LLMs and MLLMs, and highlighting the unique advantages of these LLMs and MLLMs in medical applications. Next, the survey reviews existing medical LLMs and MLLMs, providing detailed guidance on their construction and evaluation in a clear and systematic manner. Subsequently, to underscore the substantial value of LLMs and MLLMs in healthcare, the survey explores five promising applications in the field. Finally, the survey addresses the challenges confronting medical LLMs and MLLMs and proposes practical strategies and future directions for their integration into medicine. In summary, this survey offers a comprehensive analysis of the technical methodologies and practical clinical applications of medical LLMs and MLLMs, with the goal of bridging the gap between these advanced technologies and clinical practice, thereby fostering the evolution of the next generation of intelligent healthcare systems.

Zhan Peng, Xinyi Ye, Weiyue Zhao et al.

In this paper, we present RStab, a novel framework for video stabilization that integrates 3D multi-frame fusion through volume rendering. Departing from conventional methods, we introduce a 3D multi-frame perspective to generate stabilized images, addressing the challenge of full-frame generation while preserving structure. The core of our approach lies in Stabilized Rendering (SR), a volume rendering module, which extends beyond the image fusion by incorporating feature fusion. The core of our RStab framework lies in Stabilized Rendering (SR), a volume rendering module, fusing multi-frame information in 3D space. Specifically, SR involves warping features and colors from multiple frames by projection, fusing them into descriptors to render the stabilized image. However, the precision of warped information depends on the projection accuracy, a factor significantly influenced by dynamic regions. In response, we introduce the Adaptive Ray Range (ARR) module to integrate depth priors, adaptively defining the sampling range for the projection process. Additionally, we propose Color Correction (CC) assisting geometric constraints with optical flow for accurate color aggregation. Thanks to the three modules, our RStab demonstrates superior performance compared with previous stabilizers in the field of view (FOV), image quality, and video stability across various datasets.

Rongzhi Zhang, Jiaming Shen, Tianqi Liu et al.

Large Language Models (LLMs) have exhibited impressive capabilities in various tasks, yet their vast parameter sizes restrict their applicability in resource-constrained settings. Knowledge distillation (KD) offers a viable solution by transferring expertise from large teacher models to compact student models. However, traditional KD techniques face specific challenges when applied to LLMs, including restricted access to LLM outputs, significant teacher-student capacity gaps, and the inherited mis-calibration issue. In this work, we present PLaD, a novel preference-based LLM distillation framework. PLaD exploits the teacher-student capacity discrepancy to generate pseudo-preference pairs where teacher outputs are preferred over student outputs. Then, PLaD leverages a ranking loss to re-calibrate student's estimation of sequence likelihood, which steers the student's focus towards understanding the relative quality of outputs instead of simply imitating the teacher. PLaD bypasses the need for access to teacher LLM's internal states, tackles the student's expressivity limitations, and mitigates the student mis-calibration issue. Through extensive experiments on two sequence generation tasks and with various LLMs, we demonstrate the effectiveness of our proposed PLaD framework.

Yao Zhao, Rishabh Joshi, Tianqi Liu et al.

Learning from human feedback has been shown to be effective at aligning language models with human preferences. Past work has often relied on Reinforcement Learning from Human Feedback (RLHF), which optimizes the language model using reward scores assigned from a reward model trained on human preference data. In this work we show how the recently introduced Sequence Likelihood Calibration (SLiC), can also be used to effectively learn from human preferences (SLiC-HF). Furthermore, we demonstrate this can be done with human feedback data collected for a different model, similar to off-policy, offline RL data. Automatic and human evaluation experiments on the TL;DR summarization task show that SLiC-HF significantly improves supervised fine-tuning baselines. Furthermore, SLiC-HF presents a competitive alternative to the PPO RLHF implementation used in past work while being much simpler to implement, easier to tune and more computationally efficient in practice.

Jialu Liu, Tianqi Liu, Cong Yu

Effectively modeling text-rich fresh content such as news articles at document-level is a challenging problem. To ensure a content-based model generalize well to a broad range of applications, it is critical to have a training dataset that is large beyond the scale of human labels while achieving desired quality. In this work, we address those two challenges by proposing a novel approach to mine semantically-relevant fresh documents, and their topic labels, with little human supervision. Meanwhile, we design a multitask model called NewsEmbed that alternatively trains a contrastive learning with a multi-label classification to derive a universal document encoder. We show that the proposed approach can provide billions of high quality organic training examples and can be naturally extended to multilingual setting where texts in different languages are encoded in the same semantic space. We experimentally demonstrate NewsEmbed's competitive performance across multiple natural language understanding tasks, both supervised and unsupervised.

Jiaming Shen, Jialu Liu, Tianqi Liu et al.

Pre-trained text encoders such as BERT and its variants have recently achieved state-of-the-art performances on many NLP tasks. While being effective, these pre-training methods typically demand massive computation resources. To accelerate pre-training, ELECTRA trains a discriminator that predicts whether each input token is replaced by a generator. However, this new task, as a binary classification, is less semantically informative. In this study, we present a new text encoder pre-training method that improves ELECTRA based on multi-task learning. Specifically, we train the discriminator to simultaneously detect replaced tokens and select original tokens from candidate sets. We further develop two techniques to effectively combine all pre-training tasks: (1) using attention-based networks for task-specific heads, and (2) sharing bottom layers of the generator and the discriminator. Extensive experiments on GLUE and SQuAD datasets demonstrate both the effectiveness and the efficiency of our proposed method.