Step-level reward models (SRMs) can significantly enhance mathematical reasoning performance through process supervision or step-level preference alignment based on reinforcement learning. The performance of SRMs is pivotal, as they serve as critical guidelines, ensuring that each step in the reasoning process is aligned with desired outcomes. Recently, AlphaZero-like methods, where Monte Carlo Tree Search (MCTS) is employed for automatic step-level preference annotation, have proven particularly effective. However, the precise mechanisms behind the success of SRMs remain largely unexplored. To address this gap, this study delves into the counterintuitive aspects of SRMs, particularly focusing on MCTS-based approaches. Our findings reveal that the removal of natural language descriptions of thought processes has minimal impact on the efficacy of SRMs. Furthermore, we demonstrate that SRMs are adept at assessing the complex logical coherence present in mathematical language while having difficulty in natural language. These insights provide a nuanced understanding of the core elements that drive effective step-level reward modeling in mathematical reasoning. By shedding light on these mechanisms, this study offers valuable guidance for developing more efficient and streamlined SRMs, which can be achieved by focusing on the crucial parts of mathematical reasoning.

Mathematical reasoning remains a challenging area for large language models (LLMs), prompting the development of math-specific LLMs such as LLEMMA, DeepSeekMath, and Qwen2-Math, among others. These models typically follow a two-stage training paradigm: pre-training with math-related corpora and post-training with problem datasets for supervised fine-tuning (SFT). Despite these efforts, the improvements in mathematical reasoning achieved through continued pre-training (CPT) are often less significant compared to those obtained via SFT. This study addresses this discrepancy by exploring alternative strategies during the pre-training phase, focusing on the use of problem-solving data over general mathematical corpora. We investigate three primary research questions: (1) Can problem-solving data enhance the model's mathematical reasoning capabilities more effectively than general mathematical corpora during CPT? (2) Are synthetic data from the same source equally effective, and which synthesis methods are most efficient? (3) How do the capabilities developed from the same problem-solving data differ between the CPT and SFT stages, and what factors contribute to these differences? Our findings indicate that problem-solving data significantly enhances the model's mathematical capabilities compared to general mathematical corpora. We also identify effective data synthesis methods, demonstrating that the tutorship amplification synthesis method achieves the best performance. Furthermore, while SFT facilitates instruction-following abilities, it underperforms compared to CPT with the same data, which can be partially attributed to its poor learning capacity for more challenging problem-solving data. These insights provide valuable guidance for optimizing the mathematical reasoning capabilities of LLMs, culminating in our development of a powerful mathematical base model called MathGPT-8B.

Zero-shot natural language to SQL (NL2SQL) aims to generalize pretrained NL2SQL models to new environments ( e.g. , new databases and new linguistic phenomena) without any annotated NL2SQL samples from these environments. Existing approaches either use small language models (SLMs) like BART and T5, or prompt large language models (LLMs). However, SLMs may struggle with complex natural language reasoning, and LLMs may not precisely align schemas to identify the correct columns or tables. In this paper, we propose a ZeroNL2SQL framework, which divides NL2SQL into smaller sub-tasks and utilizes both SLMs and LLMs. ZeroNL2SQL first fine-tunes SLMs for better generalizability in SQL structure identification and schema alignment, producing an SQL sketch. It then uses LLMs's language reasoning capability to fill in the missing information in the SQL sketch. To support ZeroNL2SQL, we propose novel database serialization and question-aware alignment methods for effective sketch generation using SLMs. Additionally, we devise a multi-level matching strategy to recommend the most relevant values to LLMs, and select the optimal SQL query via an execution-based strategy. Comprehensive experiments show that ZeroNL2SQL achieves the best zero-shot NL2SQL performance on benchmarks, i.e. , outperforming the state-of-the-art SLM-based methods by 5.5% to 16.4% and exceeding LLM-based methods by 10% to 20% on execution accuracy.

Handwritten mathematical expression recognition (HMER) is a multidisciplinary task that generates LaTeX sequences from images.Existing approaches, employing tree decoders within attention-based encoder-decoder architectures, aim to capture the hierarchical tree structure, but are limited by CFGs and pregenerated triplet data, hindering expandability and neglecting visual ambiguity challenges.This article investigates the distinctive language characteristics of LaTeX mathematical expressions, revealing two key observations: 1) the presence of explicit structural symbols, and 2) the treatment of symbols as minimal units, each directly assigned specific semantics.Rooted in these properties, we propose that language models have the potential to synchronously and complementarily provide both structural and semantic information, making them suitable for correction of HMER.To validate our proposition, we propose an architecture called Recognition and Language Fusion Network (RLFN), which integrates recognition and language features to output corrected sequences while jointly optimizing with a string decoder recognition model.Experiments show that RLFN outperforms existing state-of-theart methods on the CROHME 2014/2016/2019 datasets.1

Data curation is a wide-ranging area which contains many critical but time-consuming data processing tasks. However, the diversity of such tasks makes it challenging to develop a general-purpose data curation system. To address this issue, we present Lingua Manga, a user-friendly and versatile system that utilizes pre-trained large language models. Lingua Manga offers automatic optimization for achieving high performance and label efficiency while facilitating flexible and rapid development. Through three example applications with distinct objectives and users of varying levels of technical proficiency, we demonstrate that Lingua Manga can effectively assist both skilled programmers and low-code or even no-code users in addressing data curation challenges.

Synthesizer is a type of electronic musical instrument that is now widely used in modern music production and sound design. Each parameters configuration of a synthesizer produces a unique timbre and can be viewed as a unique instrument. The problem of estimating a set of parameters configuration that best restore a sound timbre is an important yet complicated problem, i.e.: the synthesizer parameters estimation problem. We proposed a multi-modal deep-learning-based pipeline Sound2Synth, together with a network structure Prime-Dilated Convolution (PDC) specially designed to solve this problem. Our method achieved not only SOTA but also the first real-world applicable results on Dexed synthesizer, a popular FM synthesizer.

Despite the rising popularity of automated visualization tools, existing systems tend to provide direct results which do not always fit the input data or meet visualization requirements. Therefore, additional specification adjustments are still required in real-world use cases. However, manual adjustments are difficult since most users do not necessarily possess adequate skills or visualization knowledge. Even experienced users might create imperfect visualizations that involve chart construction errors. We present a framework, VizLinter, to help users detect flaws and rectify already-built but defective visualizations. The framework consists of two components, (1) a visualization linter, which applies well-recognized principles to inspect the legitimacy of rendered visualizations, and (2) a visualization fixer, which automatically corrects the detected violations according to the linter. We implement the framework into an online editor prototype based on Vega-Lite specifications. To further evaluate the system, we conduct an in-lab user study. The results prove its effectiveness and efficiency in identifying and fixing errors for data visualizations.

We cast a suite of information extraction tasks into a text-to-triple translation framework.Instead of solving each task relying on taskspecific datasets and models, we formalize the task as a translation between task-specific input text and output triples.By taking the taskspecific input, we enable a task-agnostic translation by leveraging the latent knowledge that a pre-trained language model has about the task.We further demonstrate that a simple pretraining task of predicting which relational information corresponds to which input text is an effective way to produce task-specific outputs.This enables the zero-shot transfer of our framework to downstream tasks.We study the zero-shot performance of this framework on open information extraction (OIE2016, NYT, WEB, PENN), relation classification (FewRel and TACRED), and factual probe (Google-RE and T-REx).The model transfers non-trivially to most tasks and is often competitive with a fully supervised method without the need for any task-specific training.For instance, we significantly outperform the F1 score of the supervised open information extraction without needing to use its training set.