ChatGLM: A Family of Large Language Models from GLM-130B to GLM-4 All Tools

We introduce ChatGLM, an evolving family of large language models that we have been developing over time. This report primarily focuses on the GLM-4 language series, which includes GLM-4, GLM-4-Air, and GLM-4-9B. They represent our most capable models that are trained with all the insights and lessons gained from the preceding three generations of ChatGLM. To date, the GLM-4 models are pre-trained on ten trillions of tokens mostly in Chinese and English, along with a small set of corpus from 24 languages, and aligned primarily for Chinese and English usage. The high-quality alignment is achieved via a multi-stage post-training process, which involves supervised fine-tuning and learning from human feedback. Evaluations show that GLM-4 1) closely rivals or outperforms GPT-4 in terms of general metrics such as MMLU, GSM8K, MATH, BBH, GPQA, and HumanEval, 2) gets close to GPT-4-Turbo in instruction following as measured by IFEval, 3) matches GPT-4 Turbo (128K) and Claude 3 for long context tasks, and 4) outperforms GPT-4 in Chinese alignments as measured by AlignBench. The GLM-4 All Tools model is further aligned to understand user intent and autonomously decide when and which tool(s) touse -- including web browser, Python interpreter, text-to-image model, and user-defined functions -- to effectively complete complex tasks. In practical applications, it matches and even surpasses GPT-4 All Tools in tasks like accessing online information via web browsing and solving math problems using Python interpreter. Over the course, we have open-sourced a series of models, including ChatGLM-6B (three generations), GLM-4-9B (128K, 1M), GLM-4V-9B, WebGLM, and CodeGeeX, attracting over 10 million downloads on Hugging face in the year 2023 alone. The open models can be accessed through https://github.com/THUDM and https://huggingface.co/THUDM.

Async Learned User Embeddings for Ads Delivery Optimization

User representation is crucial for recommendation systems as it helps to deliver personalized recommendations by capturing user preferences and behaviors in low-dimensional vectors. High-quality user embeddings can capture subtle preferences, enable precise similarity calculations, and adapt to changing preferences over time to maintain relevance. The effectiveness of recommendation systems depends significantly on the quality of user embedding. We propose to asynchronously learn high fidelity user embeddings for billions of users each day from sequence based multimodal user activities in Meta platforms through a Transformer-like large scale feature learning module. The async learned user representations embeddings (ALURE) are further converted to user similarity graphs through graph learning and then combined with user realtime activities to retrieval highly related ads candidates for the entire ads delivery system. Our method shows significant gains in both offline and online experiments.

Socially-Aware Shared Control Navigation for Assistive Mobile Robots in the Built Environment

As the number of Persons with Disabilities (PWD), particularly those with one or more physical impairments, increases, there is an increasing demand for assistive robotic technologies that can support independent mobility in the built environment and reduce the burden on caregivers. Current assistive mobility platforms (e.g., robotic wheelchairs) often fail to incorporate user preferences and control, leading to reduced trust and efficiency. Existing shared control algorithms do not allow the incorporation of the user control preferences inside the navigation framework or the path planning algorithm. In addition, existing dynamic local planner algorithms for robotic wheelchairs do not take into account the social spaces of people, potentially leading such platforms to infringe upon these areas and cause discomfort. To address these concerns, this work introduces a novel socially-aware shared autonomy-based navigation system for assistive mobile robotic platforms. Our navigation framework comprises a Global Planner and a Local Planner. To implement the Global Planner, the proposed approach introduces a novel User Preference Field (UPF) theory within its global planning framework, explicitly acknowledging user preferences to adeptly navigate away from congested areas. For the Local Planner, we propose a Socially-aware Shared Control-based Model Predictive Control with Dynamic Control Barrier Function (SS-MPC-DCBF) to adjust movements in real-time, integrating user preferences for safer, more autonomous navigation. Evaluation results show that our Global Planner aligns closely with user preferences compared to baselines, and our Local Planner demonstrates enhanced safety and efficiency in dynamic and static scenarios. This integrated approach fosters trust and autonomy, crucial for the acceptance of assistive mobility technologies in the built environment.

Libra: Building Decoupled Vision System on Large Language Models

In this work, we introduce Libra, a prototype model with a decoupled vision system on a large language model (LLM). The decoupled vision system decouples inner-modal modeling and cross-modal interaction, yielding unique visual information modeling and effective cross-modal comprehension. Libra is trained through discrete auto-regressive modeling on both vision and language inputs. Specifically, we incorporate a routed visual expert with a cross-modal bridge module into a pretrained LLM to route the vision and language flows during attention computing to enable different attention patterns in inner-modal modeling and cross-modal interaction scenarios. Experimental results demonstrate that the dedicated design of Libra achieves a strong MLLM baseline that rivals existing works in the image-to-text scenario with merely 50 million training data, providing a new perspective for future multimodal foundation models. Code is available at https://github.com/YifanXu74/Libra.

A Cause-Effect Look at Alleviating Hallucination of Knowledge-grounded Dialogue Generation

Empowered by the large-scale pretrained language models, existing dialogue systems have demonstrated impressive performance conducting fluent and natural-sounding conversations. However, they are still plagued by the hallucination problem, causing unpredictable factual errors in the generated responses. Recently, knowledge-grounded dialogue generation models, that intentionally invoke external knowledge resources to more informative responses, are also proven to be effective in reducing hallucination. Following the idea of getting high-quality knowledge, a few efforts have achieved pretty good performance on this issue. As some inevitable knowledge noises may also lead to hallucinations, it is emergent to investigate the reason and future directions for building noise-tolerant methods in KGD tasks. In this paper, we analyze the causal story behind this problem with counterfactual reasoning methods. Based on the causal effect analysis, we propose a possible solution for alleviating the hallucination in KGD by exploiting the dialogue-knowledge interaction. Experimental results of our example implementation show that this method can reduce hallucination without disrupting other dialogue performance, while keeping adaptive to different generation models. We hope our efforts can support and call for more attention to developing lightweight techniques towards robust and trusty dialogue systems.

ChatGLM-Math: Improving Math Problem-Solving in Large Language Models with a Self-Critique Pipeline

Large language models (LLMs) have shown excellent mastering of human language, but still struggle in real-world applications that require mathematical problem-solving. While many strategies and datasets to enhance LLMs' mathematics are developed, it remains a challenge to simultaneously maintain and improve both language and mathematical capabilities in deployed LLM systems.In this work, we tailor the Self-Critique pipeline, which addresses the challenge in the feedback learning stage of LLM alignment. We first train a general Math-Critique model from the LLM itself to provide feedback signals. Then, we sequentially employ rejective fine-tuning and direct preference optimization over the LLM's own generations for data collection. Based on ChatGLM3-32B, we conduct a series of experiments on both academic and our newly created challenging dataset, MathUserEval. Results show that our pipeline significantly enhances the LLM's mathematical problem-solving while still improving its language ability, outperforming LLMs that could be two times larger. Related techniques have been deployed to ChatGLM\footnote{\url{https://chatglm.cn}}, an online serving LLM. Related evaluation dataset and scripts are released at \url{https://github.com/THUDM/ChatGLM-Math}.

3DMIT: 3D Multi-modal Instruction Tuning for Scene Understanding

The remarkable potential of multi-modal large language models (MLLMs) in comprehending both vision and language information has been widely acknowledged. However, the scarcity of 3D scenes-language pairs in comparison to their 2D counterparts, coupled with the inadequacy of existing approaches in understanding of 3D scenes by LLMs, poses a significant challenge. In response, we collect and construct an extensive dataset comprising 75K instruction-response pairs tailored for 3D scenes. This dataset addresses tasks related to 3D VQA, 3D grounding, and 3D conversation. To further enhance the integration of 3D spatial information into LLMs, we introduce a novel and efficient prompt tuning paradigm, 3DMIT. This paradigm eliminates the alignment stage between 3D scenes and language and extends the instruction prompt with the 3D modality information including the entire scene and segmented objects. We evaluate the effectiveness of our method across diverse tasks in the 3D scene domain and find that our approach serves as a strategic means to enrich LLMs' comprehension of the 3D world. Our code is available at https://github.com/staymylove/3DMIT.

Forecasting Cryptocurrency Staking Rewards

This research explores a relatively unexplored area of predicting cryptocurrency staking rewards, offering potential insights to researchers and investors. We investigate two predictive methodologies: a) a straightforward sliding-window average, and b) linear regression models predicated on historical data. The findings reveal that ETH staking rewards can be forecasted with an RMSE within 0.7% and 1.1% of the mean value for 1-day and 7-day look-aheads respectively, using a 7-day sliding-window average approach. Additionally, we discern diverse prediction accuracies across various cryptocurrencies, including SOL, XTZ, ATOM, and MATIC. Linear regression is identified as superior to the moving-window average for perdicting in the short term for XTZ and ATOM. The results underscore the generally stable and predictable nature of staking rewards for most assets, with MATIC presenting a noteworthy exception.

Coordinated Navigation Control of Cross-Domain Unmanned Systems via Guiding Vector Fields

This paper proposes a distributed guiding-vector-field (DGVF) controller for cross-domain unmanned systems (CDUSs) consisting of heterogeneous unmanned aerial vehicles (UAVs) and unmanned surface vehicles (USVs), to achieve coordinated navigation whereas maneuvering along their prescribed paths. In particular, the DGVF controller provides a hierarchical architecture of an upper-level heterogeneous guidance velocity controller and a lower-level signal tracking regulator. Therein, the upper-level controller is to govern multiple heterogeneous USVs and UAVs to approach and maneuver along the prescribed paths and coordinate the formation simultaneously, whereas the low-level regulator is to track the corresponding desired guidance signals provided by the upper-level module. Significantly, the heterogeneous coordination among neighboring UAVs and USVs is achieved merely by the lightweight communication of a scalar (i.e., the additional virtual coordinate), which substantially decreases the communication and computational costs. Sufficient conditions assuring asymptotical convergence of the closed-loop system are derived in presence of the exponentially vanishing tracking errors. Finally, real-lake experiments are conducted on a self-established cross-domain heterogeneous platform consisting of three M-100 UAVs, two HUSTER-16 USVs, a HUSTER-12C USV, and a WiFi 5G wireless communication station to verify the effectiveness of the present DGVF controller.

AlignBench: Benchmarking Chinese Alignment of Large Language Models

Alignment has become a critical step for instruction-tuned Large Language Models (LLMs) to become helpful assistants. However, effective evaluation of alignment for emerging Chinese LLMs is still significantly lacking, calling for real-scenario grounded, open-ended, challenging and automatic evaluations tailored for alignment. To fill in this gap, we introduce AlignBench, a comprehensive multi-dimensional benchmark for evaluating LLMs' alignment in Chinese. Equipped with a human-in-the-loop data curation pipeline, our benchmark employs a rule-calibrated multi-dimensional LLM-as-Judge with Chain-of-Thought to generate explanations and final ratings as evaluations, ensuring high reliability and interpretability. Furthermore, we report AlignBench evaluated by CritiqueLLM, a dedicated Chinese evaluator LLM that recovers 95% of GPT-4's evaluation ability. We will provide public APIs for evaluating AlignBench with CritiqueLLM to facilitate the evaluation of LLMs' Chinese alignment. All evaluation codes, data, and LLM generations are available at \url{https://github.com/THUDM/AlignBench}.