Papers with GQA
Automatic Generation of Contrast Sets from Scene Graphs: Probing the Compositional Consistency of GQA (2021.naacl-main)
Copied to clipboard
| Challenge: | Recent studies show that supervised models exploit data artifacts to achieve good test scores while their performance severely degrades on samples outside their training distribution. |
| Approach: | They propose a method which automatically generates contrast sets for the visual question answering task by using a semantic input representation. |
| Outcome: | The proposed method computes the answer of perturbed questions, thus reducing annotation cost and enabling thorough evaluation of models’ performance on various semantic aspects. |
Multimodal Graph Transformer for Multimodal Question Answering (2023.eacl-main)
Copied to clipboard
| Challenge: | a myriad of complex tasks require both prior knowledge and reasoning intelligence. |
| Approach: | They propose a plug-and-play quasi-attention mechanism to integrate multimodal graph information to vanilla self-attention as effective prior. |
| Outcome: | The proposed model is able to perform reasoning across multiple modalities. |
MetaVL: Transferring In-Context Learning Ability From Language Models to Vision-Language Models (2023.acl-short)
Copied to clipboard
| Challenge: | Large-scale pre-trained vision-language models do not possess the ability to conduct in-context learning. |
| Approach: | They propose to meta-train a language model to perform in-context learning on NLP tasks and then transfer this model to VL tasks by attaching a visual encoder. |
| Outcome: | The proposed model outperforms the baseline model on VQA, OK-VQA, and GQA while having 20 times fewer parameters. |
Plug-and-Play VQA: Zero-shot VQA by Conjoining Large Pretrained Models with Zero Training (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Existing approaches require substantial adaptation of pretrained language models for vision-language reasoning tasks. |
| Approach: | They propose to use natural language and network interpretation as an intermediate representation that glues pretrained models together. |
| Outcome: | The proposed framework outperforms the Flamingo model on VQAv2 and GQA by 8.5%. |
WISCA: A Lightweight Model Transition Method to Improve LLM Training via Weight Scaling (2026.findings-acl)
Copied to clipboard
Jiacheng Li, Jianchao Tan, Zhidong Yang, Pingwei Sun, Feiye Huo, Jiayu Qin, Xiangyu Zhang, Maoxin He, Guangming Tan, Weile Jia, Xunliang Cai, Tong Zhao
| Challenge: | Recent advances in training optimization for Transformer-based large language models lack systematic optimization of weight patterns during training. |
| Approach: | They propose a Weight Scaling method that rescales weights while preserving model outputs to improve model training efficiency and model quality. |
| Outcome: | The proposed method significantly improves convergence quality and loss reduction in LLMs with Grouped Query Attention architectures and LoRA fine-tuning tasks. |
Recover-LoRA: Data-Free Accuracy Recovery of Degraded Language Models via Low-Rank Adaptation (2025.emnlp-industry)
Copied to clipboard
| Challenge: | Inference optimizations such as quantization, pruning, format and datatype conversion, model export, and serialization can lead to functional degradations in language model task performance. |
| Approach: | They propose a lightweight and dataset-agnostic method to recover model accuracies from quantization, pruning, format and datatype conversion, model export, and serialization errors. |
| Outcome: | The proposed method recovers model accuracies by 5-17% on MHA and GQA models. |
‘Just because you are right, doesn’t mean I am wrong’: Overcoming a bottleneck in development and evaluation of Open-Ended VQA tasks (2021.eacl-main)
Copied to clipboard
Man Luo, Shailaja Keyur Sampat, Riley Tallman, Yankai Zeng, Manuha Vancha, Akarshan Sajja, Chitta Baral
| Challenge: | Existing visual question answering datasets assume only one ground truth answer for each question. |
| Approach: | They propose alternative answer sets (AAS) of ground-truth answers to address this limitation . they modify top VQA solvers to support multiple plausible answers for a question . |
| Outcome: | The proposed approach improves on the GQA dataset and shows that it is more efficient than previous approaches. |
Cost-Optimal Grouped-Query Attention for Long-Context Modeling (2025.emnlp-main)
Copied to clipboard
Yingfa Chen, Yutong Wu, Chenyang Song, Zhen Leng Thai, Xingyu Shen, Xu Han, Zhiyuan Liu, Maosong Sun
| Challenge: | Current GQA configurations overlook how context length influences inference cost . |
| Approach: | They propose a recipe for deriving cost-optimal GQA configurations that decouple the total head size from the hidden size and allow more flexible control over attention FLOPs. |
| Outcome: | The proposed configurations reduce memory usage and FLOPs by more than 50% compared to Llama-3's GQA, with *no degradation in model capabilities*. |
GQA: Training Generalized Multi-Query Transformer Models from Multi-Head Checkpoints (2023.emnlp-main)
Copied to clipboard
| Challenge: | Multi-query attention (MQA) can lead to quality degradation and training instability . it may not be feasible to train separate models optimized for quality and inference. |
| Approach: | They propose a recipe for uptraining existing multi-head language model checkpoints into models with MQA using 5% of original training compute. |
| Outcome: | The proposed model achieves comparable quality to multi-head attention with comparable speed. |
MLKV: Multi-Layer Key-Value Heads for Memory Efficient Transformer Decoding (2025.findings-naacl)
Copied to clipboard
| Challenge: | Multi-Layer Key-Value (MLKV) sharing reduces memory usage by 6x compared to Multi-Query Attention and Grouped-Query Attributes. |
| Approach: | They propose a novel approach that extends KV sharing across transformer layers to reduce memory usage beyond what was possible with Multi-Query Attention and Grouped-Query Attributes. |
| Outcome: | The proposed approach reduces KV cache size by 6x with minimal performance loss and scales linearly with model size, batch size, and sequence length. |
Visual–Linguistic Abductive Reasoning with LLMs for Knowledge-based Visual Question Answering (2026.findings-eacl)
Copied to clipboard
| Challenge: | Recent efforts to leverage large language models for reasoning focus on visual perception and language reasoning as separate processes. |
| Approach: | They propose a method that integrates visual and linguistic modalities into interpretable abductive reasoning chains. |
| Outcome: | The proposed method improves performance on AOKVQA, OKVQA and GQA by 2.31% . it uses fuzzy scoring to select the most coherent combination, enabling unified reasoning . |
Generalization Differences between End-to-End and Neuro-Symbolic Vision-Language Reasoning Systems (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Widely used multi-modal pretrained models generalize poorly to out-of-distribution data, suggesting shortcomings in the VLE2E pipeline. |
| Approach: | They develop a segment-combine test for multi-image queries and contrast set for cross-benchmark transfer. |
| Outcome: | The proposed method shows that it is possible to train both neural and neuro-symbolic models in the same way. |
Align Attention Heads Before Merging Them: An Effective Way for Converting MHA to GQA (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Large language models (LLMs) have demonstrated exceptional performance across diverse natural language processing tasks. |
| Approach: | They propose a method for converting multi-head attention into grouped-query attention with any compression ratio of KV heads. |
| Outcome: | The proposed method can compress up to 87.5% KV heads of LLaMA2-7B model and 75% Kv heads of Sheared-LLa MA-1.3B with acceptable performance degradation. |
PropTest: Automatic Property Testing for Improved Visual Programming (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Visual Programming is an alternative to end-to-end black-box visual reasoning models. |
| Approach: | They propose a visual programming strategy that leverages Large Language Models to generate the logic of a program in the form of its source code. |
| Outcome: | The proposed method improves ViperGPT on visual question answering and referring expression comprehension with an LLM. |
LXMERT: Learning Cross-Modality Encoder Representations from Transformers (D19-1)
Copied to clipboard
| Challenge: | Existing models with better representations of visual content and language have been developed for visual-content understanding. |
| Approach: | They propose a framework to learn vision-and-language connections from Transformers models . they pre-train a large-scale Transformer model with large amounts of image-and sentence pairs . |
| Outcome: | The proposed model improves state-of-the-art on two visual-reasoning tasks by 22% . the proposed model is based on a large-scale Transformer model with three encoders . |
II-MMR: Identifying and Improving Multi-modal Multi-hop Reasoning in Visual Question Answering (2024.findings-acl)
Copied to clipboard
| Challenge: | Existing studies have focused on assessing the model’s overall accuracy without evaluating it on different reasoning cases. |
| Approach: | They propose a novel idea to identify and improve multi-modal multi-hop reasoning in VQA by using two new language prompts to find a reasoning path to reach its answer. |
| Outcome: | The proposed model improves multi-modal multi-hop reasoning in visual question answering (VQA) it finds that the proposed model is easy to answer, simply demanding “single-hop” reasoning, whereas only a few questions require “multi-hop.” |
Towards One-to-Many Visual Question Answering (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Existing Visual Question Answering systems are constrained to support domain-specific questions . a model trained on a single specific domain may not be competent for real-world application. |
| Approach: | They propose a task to enable a single model to answer as many different domains of questions as possible . they break the task down into the integration of three key abilities . |
| Outcome: | The proposed model can answer as many domains of questions as possible, the authors argue . the proposed model generalizes well to three extra zero-shot datasets, and the results are published. |
Mixture of Weight-shared Heterogeneous Group Attention Experts for Dynamic Token-wise KV Optimization (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for token-level KV optimization and grouping of tokens are inefficient and strain compute and storage resources. |
| Approach: | They propose a mixture-of-expert approach that dynamically optimizes token-wise computation and memory allocation by a token-based expert-choice routing mechanism guided by learned importance scores. |
| Outcome: | The proposed approach retains all tokens while adaptively routing them to specialized experts with varying KV group sizes, balancing granularity and efficiency. |
Latent-Condensed Transformer for Efficient Long Context Modeling (2026.acl-long)
Copied to clipboard
| Challenge: | Existing approaches address these bottlenecks separately: Multi-head Latent Attention (MLA) reduces the KV cache by projecting tokens into a low-dimensional latent space, while sparse attention reduces computation. |
| Approach: | They propose a Latent-Condensed Attention mechanism that performs structured context condensation directly within MLA's latent space. |
| Outcome: | The proposed approach reduces KV cache size and attention cost without adding parameters. |
SpindleKV: A Novel KV Cache Reduction Method Balancing Both Shallow and Deep Layers (2025.acl-long)
Copied to clipboard
| Challenge: | Large language models (LLMs) have impressive capabilities across various fields, but their widespread use is facing a severe and realistic challenge, which is their high demand for GPU memory. |
| Approach: | They propose a KV cache reduction method which balances both shallow and deep layers by using an attention weight based eviction method and a codebook based replacement approach. |
| Outcome: | The proposed method reduces the KV cache for shallower layers while preserving similar or even better model performance. |
Towards Economical Inference: Enabling DeepSeek’s Multi-Head Latent Attention in Any Transformer-based LLMs (2025.acl-long)
Copied to clipboard
Tao Ji, Bin Guo, Yuanbin Wu, Qipeng Guo, Shenlixing Shenlixing, Chenzhan Chenzhan, Xipeng Qiu, Qi Zhang, Tao Gui
| Challenge: | Multi-head Latent Attention (MLA) is an innovative architecture designed to ensure efficient and economical inference by significantly compressing the Key-Value (KV) cache into a latent vector. |
| Approach: | They propose a data-efficient fine-tuning method for transitioning from MHA to MLA using a latent vector cache. |
| Outcome: | The proposed architecture reduces the KV cache size of Llama2-7B by 92.19%, with only 1% drop in LongBench performance. |
CEBC: Conformal Evidence-Bounded Control for Low-Hallucination Vision–Language Generation (2026.acl-long)
Copied to clipboard
| Challenge: | Existing mitigation approaches reduce hallucinated object mentions at the cost of degraded generation quality or require expensive retraining and task-specific supervision. |
| Approach: | They propose a lightweight framework for low-hallucination vision–language generation . it uses evidence-bounded minimal editing to revise or suppress unsupported referenced entities . |
| Outcome: | The proposed framework reduces hallucinations while maintaining or improving quality metrics. |