Papers with 7B model
Learning Planning-based Reasoning by Trajectories Collection and Process Reward Synthesizing (2024.emnlp-main)
Copied to clipboard
| Challenge: | Recent studies have raised concerns regarding the hallucination and flaws in their reasoning process. |
| Approach: | They propose a framework to learn planning-based reasoning through Direct Preference Optimization on collected trajectories, which are ranked according to synthesized process rewards. |
| Outcome: | The proposed model surpasses GPT-3.5-Turbo on logical reasoning benchmarks on a set of logically-based reasoning tasks. |
Intent-Driven Semantic ID Generation for Grounded Conversational News Recommendation (2026.acl-industry)
Copied to clipboard
| Challenge: | a new approach to news recommendation grounds each suggestion in a rapidly evolving article corpus while addressing implicit user intents that lack explicit retrievable keywords. |
| Approach: | They propose an intent-driven Semantic ID generation paradigm to address these challenges . they map diverse intents to hierarchical SID prefixes and then fuzzy-match them to current news pool . |
| Outcome: | The proposed model achieves 0% hallucination and 12.4% L1 match on a mainstream Chinese news platform. |
PFDial: A Structured Dialogue Instruction Fine-tuning Method Based on UML Flowcharts (2025.findings-acl)
Copied to clipboard
Ming Zhang, Yuhui Wang, Yujiong Shen, Tingyi Yang, Changhao Jiang, Yilong Wu, Shihan Dou, Qinhao Chen, Zhiheng Xi, Zhihao Zhang, Yi Dong, Zhen Wang, Zhihui Fei, Mingyang Wan, Tao Liang, Guojun Ma, Qi Zhang, Tao Gui, Xuanjing Huang
| Challenge: | Large Language Models (LLMs) have shown remarkable progress in dialogue and reasoning, but they struggle to solve strictly constrained dialogue tasks. |
| Approach: | They construct a dataset that contains 12,705 high-quality Chinese dialogue instructions from 440 flowcharts containing 5,055 process nodes. |
| Outcome: | The proposed model outperforms GPT-4o models on backward transitions and outperformed GPT-42 models on the same dataset. |
PVPO: Pre-Estimated Value-Based Policy Optimization for Agentic Reasoning (2026.findings-acl)
Copied to clipboard
| Challenge: | extending grouping-based methods to agentic reasoning presents unique challenges . frequent environment interactions and tool invocations render intra-group advantage estimation unstable . |
| Approach: | They propose a grouping-based method that uses a single round of rollouts to stabilize advantage estimation. |
| Outcome: | a new RL framework outperforms grouping-based methods in retrieval tasks and advanced mathematical reasoning benchmarks. |
Reformulating Domain Adaptation of Large Language Models as Adapt-Retrieve-Revise: A Case Study on Chinese Legal Domain (2024.findings-acl)
Copied to clipboard
| Challenge: | Recent large language models like GPT-4 have demonstrated astonishing zero-shot capabilities in general domain tasks, but they often generate content with hallucinations in specific domains such as Chinese law. |
| Approach: | They propose a framework for adapting large language models (LLMs) to Chinese legal domains by reformulating generation as an adapt-retrieve-revise process. |
| Outcome: | The proposed framework outperforms existing models in the Chinese legal domain by +33.6 points in the zero-shot setting. |
Evolutionary Contrastive Distillation for Language Model Alignment (2024.findings-emnlp)
Copied to clipboard
Julian Katz-Samuels, Zheng Li, Hyokun Yun, Priyanka Nigam, Yi Xu, Vaclav Petricek, Bing Yin, Trishul Chilimbi
| Challenge: | Existing studies indicate that large language models struggle with challenging instructions. |
| Approach: | They propose a method for generating high-quality synthetic preference data to enhance the complex instruction-following capability of language models. |
| Outcome: | The proposed method exceeds the performance of current SOTA 7B models and is competitive even with open-source 70B models. |
ACECODER: Acing Coder RL via Automated Test-Case Synthesis (2025.acl-long)
Copied to clipboard
| Challenge: | Recent coder models have been driven by supervised fine-tuning, but the potential of reinforcement learning remains unexplored due to the lack of reliable reward data/model in the code domain. |
| Approach: | They propose a pipeline that generates extensive test-case pairs from existing code data and constructs preference pairs based on pass rates over sampled programs. |
| Outcome: | The proposed pipeline generates extensive (question, test-cases) pairs from existing code data and trains them with Bradley-Terry loss. |
ExecVerify: White-Box RL with Verifiable Stepwise Rewards for Code Execution Reasoning (2026.acl-long)
Copied to clipboard
| Challenge: | Existing methods for code execution reasoning are limited by the difficulty of the training data. |
| Approach: | They propose a model that uses reinforcement learning to reward correct answers from execution traces. |
| Outcome: | The proposed model improves pass@1 by up to 5.9% on code generation tasks over strong baselines. |
Leveraging Web-Crawled Data for High-Quality Fine-Tuning (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Currently, large language models are fine-tuned using expensive human-annotated data or GPT-4 generated data. |
| Approach: | They propose to use web-crawled data to train a language model on a smaller set of data . their results show that the model can convert web data with irregular formats into high-quality ones . |
| Outcome: | The proposed model outperforms open-source models larger than 32B and outperformed open-sourced models such as GPT-3.5. |
Margin Matching Preference Optimization: Enhanced Model Alignment with Granular Feedback (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods for large language models rely on binary labels that fail to capture the subtle differences in relative quality between pairs. |
| Approach: | They propose a method that incorporates relative quality margins into optimization to improve LLM policies and reward models. |
| Outcome: | The proposed approach outperforms baseline methods on popular benchmarks including MT-bench and RewardBench. |
Governance in Motion: Co-evolution of Constitutions and AI models for Scalable Safety (2025.emnlp-main)
Copied to clipboard
Chenhao Huang, Ziyu Shen, Yicong Ren, Huiyuan Zheng, Jiazheng Zhang, Mingxu Chai, Ming Zhang, Shihan Dou, Fan Mo, Jie Shi, Tao Gui, Qi Zhang, Xuanjing Huang
| Challenge: | Existing approaches to align large language models with human preferences lack flexibility . static alignment preferences lack the ability to correct misaligned behaviors as they emerge . |
| Approach: | They propose a framework that enables dynamic and continuous alignment of large language models with human preferences. |
| Outcome: | The proposed framework improves safety and accuracy of a 7B model with human annotations. |
BalanceSFT: Improving LLM Function Calling with Balanced Training Signals and Data Hardness (2026.findings-acl)
Copied to clipboard
Bingguang Hao, Zengzhuang Xu, Maolin Wang, Yuntao Wen, Yicheng Chen, Cunyin Peng, Long Chen, Xiangyu Zhao, Jinjie Gu, Chenyi Zhuang, Ji Zhang
| Challenge: | Currently, Supervised Fine-Tuning (SFT) is the prevailing method for equipping Large Language Models (LLMs) with function calling capabilities, but its effectiveness is often compromised by two challenges: 1) lengthy Chain-of-Thought (CoT) reasoning tokens dominate training signals over concise function calls in the learning objective; 2) scarcity of hard training examples. |
| Approach: | They propose a framework that uses a self-adjusted signal balancing loss and a hard data re-sampling strategy to selectively generate new, high-quality complex data guided by model errors. |
| Outcome: | The proposed framework surpasses state-of-the-art models like GPT-5 in function calling performance. |
Large Language Models are In-context Teachers for Knowledge Reasoning (2024.findings-emnlp)
Copied to clipboard
| Challenge: | In-context teaching is a method of providing in-concept example rationales to a student to reason over unseen cases. |
| Approach: | They propose to use an LLM's self-elicited explanations as in-context demonstrations to prompt a student to reason over unseen cases. |
| Outcome: | The proposed model outperforms human-crafted demonstrations on medical question answering and human-created models outperfect human-made demonstrations. |
War of Thoughts: Competition Stimulates Stronger Reasoning in Large Language Models (2025.findings-acl)
Copied to clipboard
| Challenge: | Recent advances in Large Language Models (LLMs) have reshaped the landscape of reasoning tasks. |
| Approach: | They propose a method that enhances LLM reasoning without finetuning by using test-time scaling. |
| Outcome: | The proposed method outperforms baseline models in both budget and model size. |
PARIF: Pushing the Pareto Frontier of Instruction Following and Reasoning with Curriculum Reinforcement Learning (2026.acl-long)
Copied to clipboard
| Challenge: | Existing alignment methods struggle to balance general reasoning with instruction-following (IF) this is hindered by dependency on teacher models, reward hacking, and reasoning-answer inconsistencies. |
| Approach: | They propose a two-stage curriculum learning framework based on Reinforcement Learning from Verifiable Rewards to enhance both IF and general reasoning capabilities. |
| Outcome: | The proposed framework outperforms leading models on six representative IF tasks while achieving a 21.25% relative average improvement over the original model. |
CharTide: Data-Centric Chart-to-Code Generation via Tri-Perspective Tuning and Inquiry-Driven Evolution (2026.acl-long)
Copied to clipboard
Xiangxi Zheng, Kuang He, Jiayi Hu, Ping Yu, Rui Yan, Yuan Yao, Peng Hou, Anxiang Zeng, Alex Jinpeng Wang
| Challenge: | Existing approaches to chart-to-code generation are constrained by data-centric limitations . authors present a new framework that redesigns both training and alignment data . |
| Approach: | They propose a data-centric framework that redesigns both training and alignment data for chart-to-code generation. |
| Outcome: | The proposed framework outperforms open-source baselines and is competitive with GPT-5. |
Arctic-Text2SQL-R1: Simple Rewards, Strong Reasoning in Text-to-SQL (2026.findings-acl)
Copied to clipboard
Zhewei Yao, Guoheng Sun, Łukasz Borchmann, Zheyu Shen, Minghang Deng, Bohan Zhai, Hao Zhang, Ang Li, Yuxiong He
| Challenge: | Translating natural language questions into SQL is a core challenge in natural language understanding and human-computer interaction. |
| Approach: | They propose a reinforcement learning framework and model family to generate accurate, executable SQL using a lightweight reward signal based solely on execution correctness. |
| Outcome: | The proposed framework outperforms previous versions of 70B-class systems and achieves state-of-the-art execution accuracy across six diverse Text2SQL benchmarks. |
SudokuFill: A Multi-Agent Progressive Filling Framework for Document-Level Scientific Information Extraction (2026.findings-acl)
Copied to clipboard
Yang Li, Yajiao Wang, Yu Zhang, Yuanzhe Zhang, Maodi Hu, Mengting Zhang, Xi Sun, Hua Yue, Zhixiong Zhang
| Challenge: | Scientific information extraction (SciIE) is a key bottleneck for turning unstructured papers into computable knowledge bases. |
| Approach: | They propose a scientific information extraction framework that solves a Sudoku problem as a progressive filling problem. |
| Outcome: | The proposed framework outperforms the GPT-4o model on a document-level adjuvant dataset. |
Empowering GUI Agents via Autonomous Experience Exploration and Hindsight Experience Utilization for Task Planning (2026.acl-long)
Copied to clipboard
| Challenge: | Multimodal web agents are cost-efficient and privacy-preserving, but suffer from weak planning and limited cross-website generalization. |
| Approach: | They propose a method which autonomously explores environments to discover experiences and utilizes hindsight experience to synthesize strictly aligned, high-level training data. |
| Outcome: | The proposed method outperforms Qwen2.5-VL-32B model on real-world benchmarks and demonstrates that mastering low-level atomic skills does not guarantee high-level planning competence. |
ReCode: Reinforcing Code Generation with Reasoning-Process Rewards (2026.acl-long)
Copied to clipboard
| Challenge: | Bringing process-level supervision into RL often neglects optimizing reasoning quality. |
| Approach: | They propose a framework for RL that integrates reasoning-process rewards with strict execution outcomes and a benchmark comprising preference pairs of superior and inferior reasoning processes. |
| Outcome: | The proposed framework outperforms the base version of ReCode by 16.1% and reaches performance comparable to GPT-4-Turbo. |