Evaluating Open Language Models Across Task Types, Application Domains, and Reasoning Types: An In-Depth Experimental Analysis

The rapid rise of Language Models (LMs) has expanded their use in several applications. Yet, due to constraints of model size, associated cost, or proprietary restrictions, utilizing state-of-the-art (SOTA) LLMs is not always feasible. With open, smaller LMs emerging, more applications can leverage their capabilities, but selecting the right LM can be challenging. This work conducts an in-depth experimental analysis of the semantic correctness of outputs of 10 smaller, open LMs across three aspects: task types, application domains and reasoning types, using diverse prompt styles. We demonstrate that most effective models and prompt styles vary depending on the specific requirements. Our analysis provides a comparative assessment of LMs and prompt styles using a proposed three-tier schema of aspects for their strategic selection based on use-case and other constraints. We also show that if utilized appropriately, these LMs can compete with, and sometimes outperform, SOTA LLMs like DeepSeek-v2, GPT-3.5-Turbo, and GPT-4o.

Dubo-SQL: Diverse Retrieval-Augmented Generation and Fine Tuning for Text-to-SQL

The current state-of-the-art (SOTA) for automated text-to-SQL still falls well short of expert human performance as measured by execution accuracy (EX) on the BIRD-SQL benchmark. The most accurate methods are also slow and expensive. To advance the SOTA for text-to-SQL while reducing cost and improving speed, we explore the combination of low-cost fine tuning, novel methods for diverse retrieval-augmented generation (RAG) and new input and output formats that help large language models (LLMs) achieve higher EX. We introduce two new methods, Dubo-SQL v1 and v2. Dubo-SQL v1 sets a new record for EX on the holdout test set of BIRD-SQL. Dubo-SQL v2 achieves even higher performance on the BIRD-SQL dev set. Dubo-SQL v1 relies on LLMs from OpenAI, but uses the low-cost GPT-3.5 Turbo while exceeding the performance of the next-best model using OpenAI, which instead uses the more expensive GPT-4. Dubo-SQL v1 exceeds the performance of the next-best model using GPT-3.5 by over 20%. Dubo-SQL v2 uses GPT-4 Turbo and RAG in place of fine tuning to push EX higher.

Agentic LLMs for Question Answering over Tabular Data

Question Answering over Tabular Data (Table QA) presents unique challenges due to the diverse structure, size, and data types of real-world tables. The SemEval 2025 Task 8 (DataBench) introduced a benchmark composed of large-scale, domain-diverse datasets to evaluate the ability of models to accurately answer structured queries. We propose a Natural Language to SQL (NL-to-SQL) approach leveraging large language models (LLMs) such as GPT-4o, GPT-4o-mini, and DeepSeek v2:16b to generate SQL queries dynamically. Our system follows a multi-stage pipeline involving example selection, SQL query generation, answer extraction, verification, and iterative refinement. Experiments demonstrate the effectiveness of our approach, achieving 70.5\% accuracy on DataBench QA and 71.6\% on DataBench Lite QA, significantly surpassing baseline scores of 26\% and 27\% respectively. This paper details our methodology, experimental results, and alternative approaches, providing insights into the strengths and limitations of LLM-driven Table QA.

Concise Geometric Description as a Bridge: Unleashing the Potential of LLM for Plane Geometry Problem Solving

Plane Geometry Problem Solving (PGPS) is a multimodal reasoning task that aims to solve a plane geometric problem based on a geometric diagram and problem textual descriptions. Although Large Language Models (LLMs) possess strong reasoning skills, their direct application to PGPS is hindered by their inability to process visual diagrams. Existing works typically fine-tune Multimodal LLMs (MLLMs) end-to-end on large-scale PGPS data to enhance visual understanding and reasoning simultaneously. However, such joint optimization may compromise base LLMs' inherent reasoning capability. In this work, we observe that LLM itself is potentially a powerful PGPS solver when appropriately formulating visual information as textual descriptions. We propose to train a MLLM Interpreter to generate geometric descriptions for the visual diagram, and an off-the-shelf LLM is utilized to perform reasoning. Specifically, we choose Conditional Declaration Language (CDL) as the geometric description as its conciseness eases the MLLM Interpreter training. The MLLM Interpreter is fine-tuned via CoT (Chain-of-Thought)-augmented SFT followed by GRPO to generate CDL. Instead of using a conventional solution-based reward that compares the reasoning result with the ground-truth answer, we design CDL matching rewards to facilitate more effective GRPO training, which provides more direct and denser guidance for CDL generation. To support training, we construct a new dataset, Formalgeo7k-Rec-CoT, by manually reviewing Formalgeo7k v2 and incorporating CoT annotations. Extensive experiments on Formalgeo7k-Rec-CoT, Unigeo, and MathVista show our method (finetuned on only 5.5k data) performs favorably against leading open-source and closed-source MLLMs.

Bidirectional Awareness Induction in Autoregressive Seq2Seq Models

Autoregressive Sequence-To-Sequence models are the foundation of many Deep Learning achievements in major research fields such as Vision and Natural Language Processing. Despite that, they still present significant limitations. For instance, when errors occur in the early steps of the prediction, the whole output is severely affected. Such reliance on previously predicted tokens and the inherent computational unfriendliness of sequential algorithms, motivated researchers to explore different architectures and methods in the search for bidirectional approaches. In this work, we introduce the Bidirectional Awareness Induction (BAI), a training method that leverages a subset of elements in the network, the Pivots, to perform bidirectional learning without breaking the autoregressive constraints. To showcase its flexibility, we apply the method to three architectures, the Transformer, ExpansionNet v2 and GPT, then perform experiments over three tasks. Experimental results showcase BAI's effectiveness on all selected tasks and architectures. In particular, we observed an increase of up to 2.4 CIDEr in Image-Captioning, 4.96 BLEU in Neural Machine Translation, and 1.16 ROUGE in Text Summarization compared to the respective baselines. Notably, BAI not only has a positive impact on models trained from scratch but on pre-trained models as well. Such an aspect, combined with the absence of architectural requirements synergizes well with the current trend of LLMs.

ReaderLM-v2: Small Language Model for HTML to Markdown and JSON

We present ReaderLM-v2, a compact 1.5 billion parameter language model designed for efficient web content extraction. Our model processes documents up to 512K tokens, transforming messy HTML into clean Markdown or JSON formats with high accuracy -- making it an ideal tool for grounding large language models. The model's effectiveness results from two key innovations: (1) a three-stage data synthesis pipeline that generates high quality, diverse training data by iteratively drafting, refining, and critiquing web content extraction; and (2) a unified training framework combining continuous pre-training with multi-objective optimization. Intensive evaluation demonstrates that ReaderLM-v2 outperforms GPT-4o-2024-08-06 and other larger models by 15-20\% on carefully curated benchmarks, particularly excelling at documents exceeding 100K tokens, while maintaining significantly lower computational requirements.

AnaloBench: Benchmarking the Identification of Abstract and Long-context Analogies

Humans regularly engage in analogical thinking, relating personal experiences to current situations ($X$ is analogous to $Y$ because of $Z$). Analogical thinking allows humans to solve problems in creative ways, grasp difficult concepts, and articulate ideas more effectively. Can language models (LMs) do the same? To answer this question, we propose ANALOBENCH, a benchmark to determine analogical reasoning ability in LMs. Our benchmarking approach focuses on aspects of this ability that are common among humans: (i) recalling related experiences from a large amount of information, and (ii) applying analogical reasoning to complex and lengthy scenarios. We test a broad collection of proprietary models (e.g., GPT family, Claude V2) and open source models such as LLaMA2. As in prior results, scaling up LMs results in some performance boosts. Surprisingly, scale offers minimal gains when, (i) analogies involve lengthy scenarios, or (ii) recalling relevant scenarios from a large pool of information, a process analogous to finding a needle in a haystack. We hope these observations encourage further research in this field.

The All-Seeing Project V2: Towards General Relation Comprehension of the Open World

We present the All-Seeing Project V2: a new model and dataset designed for understanding object relations in images. Specifically, we propose the All-Seeing Model V2 (ASMv2) that integrates the formulation of text generation, object localization, and relation comprehension into a relation conversation (ReC) task. Leveraging this unified task, our model excels not only in perceiving and recognizing all objects within the image but also in grasping the intricate relation graph between them, diminishing the relation hallucination often encountered by Multi-modal Large Language Models (MLLMs). To facilitate training and evaluation of MLLMs in relation understanding, we created the first high-quality ReC dataset ({AS-V2) which is aligned with the format of standard instruction tuning data. In addition, we design a new benchmark, termed Circular-based Relation Probing Evaluation (CRPE) for comprehensively evaluating the relation comprehension capabilities of MLLMs. Notably, our ASMv2 achieves an overall accuracy of 52.04 on this relation-aware benchmark, surpassing the 43.14 of LLaVA-1.5 by a large margin. We hope that our work can inspire more future research and contribute to the evolution towards artificial general intelligence. Our project is released at https://github.com/OpenGVLab/all-seeing.

V-RECS, a Low-Cost LLM4VIS Recommender with Explanations, Captioning and Suggestions

NL2VIS (natural language to visualization) is a promising and recent research area that involves interpreting natural language queries and translating them into visualizations that accurately represent the underlying data. As we navigate the era of big data, NL2VIS holds considerable application potential since it greatly facilitates data exploration by non-expert users. Following the increasingly widespread usage of generative AI in NL2VIS applications, in this paper we present V-RECS, the first LLM-based Visual Recommender augmented with explanations(E), captioning(C), and suggestions(S) for further data exploration. V-RECS' visualization narratives facilitate both response verification and data exploration by non-expert users. Furthermore, our proposed solution mitigates computational, controllability, and cost issues associated with using powerful LLMs by leveraging a methodology to effectively fine-tune small models. To generate insightful visualization narratives, we use Chain-of-Thoughts (CoT), a prompt engineering technique to help LLM identify and generate the logical steps to produce a correct answer. Since CoT is reported to perform poorly with small LLMs, we adopted a strategy in which a large LLM (GPT-4), acting as a Teacher, generates CoT-based instructions to fine-tune a small model, Llama-2-7B, which plays the role of a Student. Extensive experiments-based on a framework for the quantitative evaluation of AI-based visualizations and on manual assessment by a group of participants-show that V-RECS achieves performance scores comparable to GPT-4, at a much lower cost. The efficacy of the V-RECS teacher-student paradigm is also demonstrated by the fact that the un-tuned Llama fails to perform the task in the vast majority of test cases. We release V-RECS for the visualization community to assist visualization designers throughout the entire visualization generation process.

TOGGLE: Temporal Logic-Guided Large Language Model Compression for Edge

Large Language Models (LLMs) deliver exceptional performance across natural language tasks but demand substantial computational resources, limiting their deployment on resource-constrained edge devices. Existing compression techniques, such as quantization and pruning, often degrade critical linguistic properties and lack formal guarantees for preserving model behavior. We propose Temporal Logic-Guided Large Language Model Compression (TOGGLE), a novel framework that leverages Signal Temporal Logic (STL) to formally specify and enforce linguistic properties during compression. TOGGLE employs an STL robustness-guided Bayesian optimization to systematically explore layer-wise quantization and pruning configurations, generating compressed models that formally satisfy specified linguistic constraints without retraining or fine-tuning. Evaluating TOGGLE on four LLM architectures (GPT-2, DeepSeek-V2 7B, LLaMA 3 8B, and Mistral 7B), we achieve up to 3.3x reduction in computational costs (FLOPs) and up to a 68.8% reduction in model size while satisfying all linguistic properties. TOGGLE represents the first integration of formal methods into LLM compression, enabling efficient, verifiable deployment of LLMs on edge hardware.