Large language models (LLMs) developed, deployed, and researched across the GCC — including Falcon (TII), Jais (MBZUAI/Inception), AceGPT (KAUST), ALLaM (SDAIA), and Fanar (QCRI).
KAUST researchers have developed a parameter-efficient learning approach to identify Arabic dialects using limited data and computing power, fine-tuning the Whisper model with a dataset of 17 dialects. The model achieves high accuracy using only 2.5% of the parameters of the larger model and 30% of the training data. Srijith Radhakrishnan presented the findings at EMNLP 2023 and Interspeech 2023. Why it matters: This research addresses the challenge of dialect identification in Arabic NLP and enables more efficient use of large language models in resource-constrained environments.
Researchers introduce ArabLegalEval, a multitask benchmark dataset for assessing Arabic legal knowledge in LLMs. The dataset contains tasks sourced from Saudi legal documents and synthesized questions, drawing inspiration from MMLU and LegalBench. Experiments benchmarked models including GPT-4 and Jais, exploring in-context learning and various evaluation methods. Why it matters: This resource should help accelerate AI research and evaluation in the Arabic legal domain, where datasets are lacking.
FancyVideo, a new video generator, introduces a Cross-frame Textual Guidance Module (CTGM) to enhance text-to-video models. CTGM uses a Temporal Information Injector and Temporal Affinity Refiner to achieve frame-specific textual guidance, improving comprehension of temporal logic. Experiments on the EvalCrafter benchmark demonstrate FancyVideo's state-of-the-art performance in generating dynamic and consistent videos, also supporting image-to-video tasks.
This paper introduces two methods for creating Arabic LLM prompts at scale: translating existing English prompt datasets and creating natural language prompts from Arabic NLP datasets. Using these methods, the authors generated over 67.4 million Arabic prompts covering tasks like summarization and question answering. Fine-tuning a 7B Qwen2 model on these prompts outperforms a 70B Llama3 model in handling Arabic prompts. Why it matters: The research provides a cost-effective approach to scaling Arabic LLM training data, potentially improving the performance of smaller, more accessible models for Arabic NLP.
MBZUAI researchers release OpenFactCheck, a unified framework to evaluate the factual accuracy of large language models. The framework includes modules for response evaluation, LLM evaluation, and fact-checker evaluation. OpenFactCheck is available as an open-source Python library, a web service, and via GitHub.
This paper introduces a nested embedding learning framework for Arabic NLP, utilizing Matryoshka Embedding Learning and multilingual models. The authors translated sentence similarity datasets into Arabic to enable comprehensive evaluation. Experiments on the Arabic Natural Language Inference dataset show Matryoshka embedding models outperform traditional models by 20-25% in capturing Arabic semantic nuances. Why it matters: This work advances Arabic NLP by providing a new method and evaluation benchmark for semantic similarity, which is crucial for tasks like information retrieval and text understanding.
The paper introduces ALLaM, a series of large language models for Arabic and English, designed to support Arabic Language Technologies. The models are trained with language alignment and knowledge transfer in mind, using a decoder-only architecture. ALLaM achieves state-of-the-art results on Arabic benchmarks like MMLU Arabic and Arabic Exams. Why it matters: This work advances Arabic NLP by providing high-performing LLMs and demonstrating effective techniques for cross-lingual transfer learning and alignment with human preferences.
The paper introduces AlcLaM, an Arabic dialectal language model trained on 3.4M sentences from social media. AlcLaM expands the vocabulary and retrains a BERT-based model, using only 13GB of dialectal text. Despite the smaller training data, AlcLaM outperforms models like CAMeL, MARBERT, and ArBERT on various Arabic NLP tasks. Why it matters: AlcLaM offers a more efficient and accurate approach to Arabic NLP by focusing on dialectal Arabic, which is often underrepresented in existing models.
The paper introduces NativQA, a language-independent framework for constructing culturally and regionally aligned QA datasets in native languages. Using the framework, the authors created MultiNativQA, a multilingual natural QA dataset consisting of ~64k manually annotated QA pairs in seven languages. The dataset covers queries from native speakers from 9 regions covering 18 topics, and is designed for evaluating and tuning LLMs. Why it matters: The framework and dataset enable the creation of more culturally relevant and effective LLMs for diverse linguistic communities, including those in the Middle East.
The paper introduces InstAr-500k, a new Arabic instruction dataset of 500,000 examples designed to improve LLM performance in Arabic. Researchers fine-tuned the open-source Gemma-7B model using InstAr-500k and evaluated it on downstream tasks, achieving strong results on Arabic NLP benchmarks. They then released GemmAr-7B-V1, a model specifically tuned for Arabic NLP tasks. Why it matters: This work addresses the lack of high-quality Arabic instruction data, potentially boosting the capabilities of Arabic language models.
MBZUAI researchers introduce Web2Code, a new large-scale dataset and evaluation framework for training and benchmarking multimodal LLMs on webpage understanding and HTML code generation. The dataset includes webpage images, HTML code, and QA pairs about webpage content. Experiments demonstrate the dataset's utility in webpage understanding, code generation, and general visual domain tasks, with code and data available on Github.
Researchers introduce two new benchmarks, derived from the Qiyas exam, to evaluate mathematical reasoning and language understanding in Arabic. They tested ChatGPT-3.5-turbo and ChatGPT-4, which achieved 49% and 64% accuracy respectively. The new benchmarks aim to address the lack of resources for evaluating Arabic language models.
MBZUAI researchers introduce VideoGPT+, a novel video Large Multimodal Model (LMM) that integrates image and video encoders to leverage both spatial and temporal information in videos. They also introduce VCGBench-Diverse, a comprehensive benchmark for evaluating video LMMs across 18 video categories. VideoGPT+ demonstrates improved performance on multiple video benchmarks, including VCGBench and MVBench.
This survey paper reviews recent literature on continual learning in medical imaging, addressing challenges like catastrophic forgetting and distribution shifts. It covers classification, segmentation, detection, and other tasks, while providing a taxonomy of studies and identifying challenges. The authors also maintain a GitHub repository to keep the survey up-to-date with the latest research.
This paper introduces SimulMask, a new paradigm for fine-tuning large language models (LLMs) for simultaneous translation. SimulMask utilizes a novel attention masking approach that models simultaneous translation during fine-tuning by masking attention for a desired decision policy. Applied to a Falcon LLM on the IWSLT 2017 dataset, SimulMask achieves improved translation quality compared to state-of-the-art prompting optimization strategies across five language pairs while reducing computational cost. Why it matters: The proposed method offers a more efficient way to adapt LLMs for real-time translation, potentially enhancing multilingual communication tools and services.
Researchers from MBZUAI have introduced the Complex Video Reasoning and Robustness Evaluation Suite (CVRR-ES) for assessing Video-LLMs. The benchmark evaluates models across 11 real-world video dimensions, revealing challenges in robustness and reasoning, particularly for open-source models. A training-free Dual-Step Contextual Prompting (DSCP) technique is proposed to enhance Video-LMM performance, with the dataset and code made publicly available.
Researchers compiled a 101 Billion Arabic Words Dataset by mining text from Common Crawl WET files and rigorously cleaning and deduplicating the extracted content. The dataset aims to address the scarcity of original, high-quality Arabic linguistic data, which often leads to bias in Arabic LLMs that rely on translated English data. This is the largest Arabic dataset available to date. Why it matters: The new dataset can significantly contribute to the development of authentic Arabic LLMs that are more linguistically and culturally accurate.
A study compared the vulnerability of C programs generated by nine state-of-the-art Large Language Models (LLMs) using a zero-shot prompt. The researchers introduced FormAI-v2, a dataset of 331,000 C programs generated by these LLMs, and found that at least 62.07% of the generated programs contained vulnerabilities, detected via formal verification. The research highlights the need for risk assessment and validation when deploying LLM-generated code in production environments.
Researchers at MBZUAI have developed DynaMMo, a dynamic model merging method for efficient class incremental learning using medical images. DynaMMo merges multiple networks at different training stages using lightweight learnable modules, reducing computational overhead. Evaluated on three datasets, DynaMMo achieved a 10-fold reduction in GFLOPS compared to existing dynamic methods with a 2.76 average accuracy drop.
Researchers introduce ArabicaQA, a large-scale dataset for Arabic question answering, comprising 89,095 answerable and 3,701 unanswerable questions. They also present AraDPR, a dense passage retrieval model trained on the Arabic Wikipedia. The paper includes benchmarking of large language models (LLMs) for Arabic question answering. Why it matters: This work addresses a significant gap in Arabic NLP resources and provides valuable tools and benchmarks for advancing research in the field.
The paper introduces MedPromptX, a clinical decision support system using multimodal large language models (MLLMs), few-shot prompting (FP), and visual grounding (VG) for chest X-ray diagnosis, integrating imagery with EHR data. MedPromptX refines few-shot data dynamically for real-time adjustment to new patient scenarios and narrows the search area in X-ray images. The study introduces MedPromptX-VQA, a new visual question answering dataset, and demonstrates state-of-the-art performance with an 11% improvement in F1-score compared to baselines.
Researchers at MBZUAI have introduced TiBiX, a novel approach leveraging temporal information from previous chest X-rays (CXRs) and reports for bidirectional generation of current CXRs and reports. TiBiX addresses two key challenges: generating current images from previous images and reports, and generating current reports from both previous and current images. The study also introduces a curated temporal benchmark dataset derived from the MIMIC-CXR dataset and achieves state-of-the-art results in report generation.
The paper introduces AraPoemBERT, an Arabic language model pretrained exclusively on 2.09 million verses of Arabic poetry. AraPoemBERT was evaluated against five other Arabic language models on tasks including poet's gender classification (99.34% accuracy) and poetry sub-meter classification (97.79% accuracy). The model achieved state-of-the-art results in these and other downstream tasks, and is publicly available on Hugging Face. Why it matters: This specialized model advances Arabic NLP by providing a new state-of-the-art tool tailored for the nuances of classical Arabic poetry.
This paper explores the impact of tokenization strategies and vocabulary sizes on Arabic language model performance across NLP tasks like news classification and sentiment analysis. It compares four tokenizers, finding that Byte Pair Encoding (BPE) with Farasa performs best overall due to its morphological analysis capabilities. The study surprisingly found limited impact of vocabulary size on performance with fixed model sizes, challenging assumptions about vocabulary size and model performance. Why it matters: The findings provide insights for developing more effective and nuanced Arabic language models, particularly for handling dialectal variations and promoting responsible AI development in the region.
The paper introduces AraTrust, a new benchmark for evaluating the trustworthiness of LLMs when prompted in Arabic. The benchmark contains 522 multiple-choice questions covering dimensions like truthfulness, ethics, safety, and fairness. Experiments using AraTrust showed that GPT-4 performed the best, while open-source models like AceGPT 7B and Jais 13B had lower scores. Why it matters: This benchmark addresses a critical gap in evaluating LLMs for Arabic, which is essential for ensuring the safe and ethical deployment of AI in the Arab world.
Researchers from MBZUAI have released MobiLlama, a fully transparent open-source 0.5 billion parameter Small Language Model (SLM). MobiLlama is designed for resource-constrained devices, emphasizing enhanced performance with reduced resource demands. The full training data pipeline, code, model weights, and checkpoints are available on Github.
The paper introduces ArabianGPT, a suite of transformer-based language models designed specifically for Arabic, including versions with 0.1B and 0.3B parameters. A key component is the AraNizer tokenizer, tailored for Arabic script's morphology. Fine-tuning ArabianGPT-0.1B achieved 95% accuracy in sentiment analysis, up from 56% in the base model, and improved F1 scores in summarization. Why it matters: The models address the gap in native Arabic LLMs, offering better performance on Arabic NLP tasks through tailored architecture and tokenization.
Researchers introduce PALO, a polyglot large multimodal model with visual reasoning capabilities in 10 major languages including Arabic. A semi-automated translation approach was used to adapt the multimodal instruction dataset from English to the target languages. The models are trained across three scales (1.7B, 7B and 13B parameters) and a multilingual multimodal benchmark is proposed for evaluation.
MBZUAI researchers introduce BiMediX, a bilingual (English and Arabic) mixture of experts LLM for medical applications. The model is trained on BiMed1.3M, a new 1.3 million bilingual instruction dataset and outperforms existing models like Med42 and Jais-30B on medical benchmarks. Code and models are available on Github.
This paper investigates the intrinsic self-correction capabilities of LLMs, identifying model confidence as a key latent factor. Researchers developed an "If-or-Else" (IoE) prompting framework to guide LLMs in assessing their own confidence and improving self-correction accuracy. Experiments demonstrate that the IoE-based prompt enhances the accuracy of self-corrected responses, with code available on GitHub.
MBZUAI researchers introduce M4GT-Bench, a new benchmark for evaluating machine-generated text (MGT) detection across multiple languages and domains. The benchmark includes tasks for binary MGT detection, identifying the specific model that generated the text, and detecting mixed human-machine text. Experiments with baseline models and human evaluation show that MGT detection performance is highly dependent on access to training data from the same domain and generators.
Researchers from the National Center for AI in Saudi Arabia investigated the sensitivity of Large Language Model (LLM) leaderboards to minor benchmark perturbations. They found that small changes, like choice order, can shift rankings by up to 8 positions. The study recommends hybrid scoring and warns against over-reliance on simple benchmark evaluations, providing code for further research.
This paper studies the impact of data scale on Arabic Pretrained Language Models (PLMs). Researchers retrained BERT-base and T5-base models on large Arabic corpora, achieving state-of-the-art results on the ALUE and ORCA benchmarks. The analysis indicates that pretraining data volume is the most important factor for performance. Why it matters: This work provides valuable insights into building effective Arabic language models, emphasizing the importance of large, high-quality datasets for advancing Arabic NLP.
Researchers introduce Arabic Mini-ClimateGPT, a tailored Arabic LLM for climate change and sustainability. The model is fine-tuned on the Clima500-Instruct dataset and uses vector embedding retrieval during inference. Evaluations show the model outperforms baseline LLMs and is preferred by experts in 81.6% of cases.
Researchers at MBZUAI have developed GeoChat, a new vision-language model (VLM) specifically designed for remote sensing imagery. GeoChat addresses the limitations of general-domain VLMs in accurately interpreting high-resolution remote sensing data, offering both image-level and region-specific dialogue capabilities. The model is trained on a novel remote sensing multimodal instruction-following dataset and demonstrates strong zero-shot performance across tasks like image captioning and visual question answering.
MBZUAI researchers introduce PG-Video-LLaVA, a large multimodal model with pixel-level grounding capabilities for videos, integrating audio cues for enhanced understanding. The model uses an off-the-shelf tracker and grounding module to localize objects in videos based on user prompts. PG-Video-LLaVA is evaluated on video question-answering and grounding benchmarks, using Vicuna instead of GPT-3.5 for reproducibility.
This paper introduces a predictive analysis of Arabic court decisions, utilizing 10,813 real commercial court cases. The study evaluates LLaMA-7b, JAIS-13b, and GPT3.5-turbo models under zero-shot, one-shot, and fine-tuned training paradigms, also experimenting with summarization and translation. GPT-3.5 models significantly outperformed others, exceeding JAIS model performance by 50%, while also demonstrating the unreliability of most automated metrics. Why it matters: This research bridges computational linguistics and Arabic legal analytics, offering insights for enhancing judicial processes and legal strategies in the Arabic-speaking world.
Researchers introduce AceGPT, a localized large language model (LLM) specifically for Arabic, addressing cultural sensitivity and local values not well-represented in mainstream models. AceGPT incorporates further pre-training with Arabic texts, supervised fine-tuning using native Arabic instructions and GPT-4 responses, and reinforcement learning with AI feedback using a reward model attuned to local culture. Evaluations demonstrate that AceGPT achieves state-of-the-art performance among open Arabic LLMs across several benchmarks. Why it matters: This work advances culturally-aware AI development for Arabic-speaking communities, providing a valuable resource and benchmark for future research.
Researchers at MBZUAI release SlimPajama-DC, an empirical analysis of data combinations for pretraining LLMs using the SlimPajama dataset. The study examines the impact of global vs. local deduplication and the proportions of highly-deduplicated multi-source datasets. Results show that increased data diversity after global deduplication is crucial, with the best configuration outperforming models trained on RedPajama.
G42's Core42 has released Jais, a collection of Arabic large language models, including a 13B parameter version. Jais-13B is trained on a 395B token dataset containing Arabic and English text. According to the blog post, Jais-13B achieves state-of-the-art results on Arabic NLP benchmarks. Why it matters: This release establishes a new benchmark for Arabic language AI, potentially enabling more sophisticated and culturally relevant applications.
G42's Core42 has released Jais, a new Arabic large language model. Jais includes 13 billion parameters and was trained on a dataset of 126B tokens, including 43B Arabic tokens. According to the developers, Jais achieves state-of-the-art results on Arabic benchmarks and competitive performance on English benchmarks. Why it matters: Jais represents a significant step forward for Arabic NLP, providing a powerful new tool for researchers and developers in the region.
MBZUAI has released Jais and Jais-chat, two new open generative large language models (LLMs) with a focus on Arabic. The 13 billion parameter models are based on the GPT-3 architecture and pretrained on Arabic, English, and code. Evaluation shows state-of-the-art Arabic knowledge and reasoning, with competitive English performance.
This paper introduces Pulmonary Embolism Detection using Contrastive Learning (PECon), a supervised contrastive pretraining strategy using both CT scans and EHR data to improve feature alignment between modalities for better PE diagnosis. PECon pulls sample features of the same class together while pushing away features of other classes. The approach achieves state-of-the-art results on the RadFusion dataset, with an F1-score of 0.913 and AUROC of 0.943.
MBZUAI researchers introduce XrayGPT, a conversational medical vision-language model for analyzing chest radiographs and answering open-ended questions. The model aligns a medical visual encoder (MedClip) with a fine-tuned large language model (Vicuna) using a linear transformation. To enhance performance, the LLM was fine-tuned using 217k interactive summaries generated from radiology reports.
Video-ChatGPT is a new multimodal model that combines a video-adapted visual encoder with a large language model (LLM) to enable detailed video understanding and conversation. The authors introduce a new dataset of 100,000 video-instruction pairs for training the model. They also develop a quantitative evaluation framework for video-based dialogue models.
The paper introduces Sparse-Quantized Representation (SpQR), a new compression format and quantization technique for large language models (LLMs). SpQR identifies outlier weights and stores them in higher precision while compressing the remaining weights to 3-4 bits. The method achieves less than 1% accuracy loss in perplexity for LLaMA and Falcon LLMs and enables a 33B parameter LLM to run on a single 24GB consumer GPU. Why it matters: This enables near-lossless compression of LLMs, making powerful models accessible on resource-constrained devices and accelerating inference without significant accuracy degradation.
This paper presents a comprehensive evaluation of ChatGPT's performance across 44 Arabic NLP tasks using over 60 datasets. The study compares ChatGPT's capabilities in Modern Standard Arabic (MSA) and Dialectal Arabic (DA) against smaller, fine-tuned models. Results show ChatGPT is outperformed by smaller, fine-tuned models and exhibits limitations in handling Arabic dialects compared to MSA. Why it matters: The work highlights the need for further research and development of Arabic-specific NLP models to overcome the limitations of general-purpose models like ChatGPT.
LAraBench introduces a benchmark for Arabic NLP and speech processing, evaluating LLMs like GPT-3.5-turbo, GPT-4, BLOOMZ, Jais-13b-chat, Whisper, and USM. The benchmark covers 33 tasks across 61 datasets, using zero-shot and few-shot learning techniques. Results show that SOTA models generally outperform LLMs in zero-shot settings, though larger LLMs with few-shot learning reduce the gap. Why it matters: This benchmark helps assess and improve the performance of LLMs on Arabic language tasks, highlighting areas where specialized models still excel.
MBZUAI researchers introduce M4, a multi-generator, multi-domain, and multi-lingual benchmark dataset for detecting machine-generated text. The study reveals challenges in generalizing detection across unseen domains or LLMs, with detectors often misclassifying machine-generated text as human-written. The dataset aims to foster research into more robust detection methods and is available on GitHub.
MBZUAI releases Bactrian-X, a multilingual parallel dataset of 3.4 million instruction-response pairs across 52 languages. They trained low-rank adaptation (LoRA) adapters using this dataset, creating lightweight, replaceable components for large language models. Experiments show the LoRA-based models outperform vanilla and existing instruction-tuned models in multilingual settings.