metaai·lightalo unofficial · independent

The papers

Every star in this catalog links to its sources. This is what those sources look like as one body of work: 74 arXiv papers, and how often the rest of the field has cited each one back. It is the shortest route from "that sounds interesting" to the actual research.

Citation counts from the Semantic Scholar Graph API, read 2026-09-04 and refreshed weekly.

74papers in the catalog
246kcitations of them, in total
38kof those judged influential
577median citations per paper
2016–2026publication years covered

Where to start

The twelve most-cited, which is a reasonable reading order for anyone new to the field. Bear in mind that a 2017 paper has had eight years to accumulate what a 2025 paper has had months for, so this ranks foundations, not importance.

Mask R-CNN 33k · 2017 RoBERTa 31k · 2019 LLaMA 21k · 2023 Llama 3 / 3.1 / 3.2 / 3.3 18k · 2024 Llama 2 18k · 2023 Segment Anything 15k · 2023 BART 13k · 2019 fastText 11k · 2016 DINOv2 10k · 2023 DINO 10k · 2021 wav2vec 2.0 9.2k · 2020 XLM-R 9.1k · 2019

All 74 papers

Sort by any column; filter by constellation or search the titles. Paper titles open on arXiv, project names open that project's page here.

74 papers
Constellation Venue
Mask R-CNN Mask R-CNN Vision 2017 32,927 4355
RoBERTa: A Robustly Optimized BERT Pretraining Approach RoBERTa Language & LLMs 2019 arXiv.org 31,147 6068
LLaMA: Open and Efficient Foundation Language Models LLaMA Language & LLMs 2023 arXiv.org 21,436 2209
The Llama 3 Herd of Models Llama 3 / 3.1 / 3.2 / 3.3 Language & LLMs 2024 18,281 3374
Llama 2: Open Foundation and Fine-Tuned Chat Models Llama 2 Language & LLMs 2023 arXiv.org 18,110 2260
Segment Anything Segment Anything Vision 2023 IEEE International Conference on C 15,338 2017
BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension BART Language & LLMs 2019 Annual Meeting of the Association 13,175 2394
Enriching Word Vectors with Subword Information fastText Open Source Infra 2016 Transactions of the Association fo 10,941 1070
DINOv2: Learning Robust Visual Features without Supervision DINOv2 Vision 2023 Trans. Mach. Learn. Res. 10,180 1470
Emerging Properties in Self-Supervised Vision Transformers DINO Vision 2021 IEEE International Conference on C 10,104 1700
wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations wav2vec 2.0 Speech & Sound 2020 Neural Information Processing Syst 9,159 3191
Unsupervised Cross-lingual Representation Learning at Scale XLM-R Language & LLMs 2019 Annual Meeting of the Association 9,113 1824
OPT: Open Pre-trained Transformer Language Models OPT-175B Language & LLMs 2022 arXiv.org 4,933 533
SAM 2: Segment Anything in Images and Videos SAM 2 Vision 2024 International Conference on Learni 4,075 591
Code Llama: Open Foundation Models for Code Code Llama Language & LLMs 2023 arXiv.org 3,531 394
fairseq: A Fast, Extensible Toolkit for Sequence Modeling fairseq Language & LLMs 2019 North American Chapter of the Asso 3,419 286
Make-A-Video: Text-to-Video Generation without Text-Video Data Make-A-Video Generative Media 2022 International Conference on Learni 2,155 110
Ego4D: Around the World in 3,000 Hours of Egocentric Video Ego4D World Models & Embodied AI 2021 Computer Vision and Pattern Recogn 2,004 286
No Language Left Behind: Scaling Human-Centered Machine Translation No Language Left Behind Speech & Sound 2022 arXiv.org 1,968 174
ImageBind One Embedding Space to Bind Them All ImageBind Vision 2023 Computer Vision and Pattern Recogn 1,754 226
VGGT: Visual Geometry Grounded Transformer VGGT Vision 2025 Computer Vision and Pattern Recogn 1,680 447
DINOv3 DINOv3 Vision 2025 1,375 240
Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversations Purple Llama Safety & Trust 2023 arXiv.org 1,286 205
Accelerating 3D deep learning with PyTorch3D PyTorch3D Vision 2019 SIGGRAPH Asia 2020 Courses 1,189 90
Self-Supervised Learning from Images with a Joint-Embedding Predictive Architecture I-JEPA World Models & Embodied AI 2023 Computer Vision and Pattern Recogn 1,175 143
GAIA: a benchmark for General AI Assistants GAIA Benchmark Safety & Trust 2023 arXiv.org 1,164 199
Galactica: A Large Language Model for Science Galactica Language & LLMs 2022 arXiv.org 1,088 97
Chameleon: Mixed-Modal Early-Fusion Foundation Models Chameleon Vision 2024 arXiv.org 956 111
The Open Catalyst 2020 (OC20) Dataset and Community Challenges Open Catalyst Project AI for Science 2020 ACS Catalysis 870 90
The Faiss Library faiss Open Source Infra 2024 IEEE Transactions on Big Data 861 53
SAM 3: Segment Anything with Concepts SAM 3 Vision 2025 arXiv.org 851 129
Simple and Controllable Music Generation AudioCraft Speech & Sound 2023 Neural Information Processing Syst 787 107
Scaling Speech Technology to 1, 000+ Languages Massively Multilingual Speech Speech & Sound 2023 Journal of machine learning resear 727 123
Training Large Language Models to Reason in a Continuous Latent Space Coconut Language & LLMs 2024 arXiv.org 705 140
V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planning V-JEPA 2 World Models & Embodied AI 2025 arXiv.org 647 96
Movie Gen: A Cast of Media Foundation Models Movie Gen Generative Media 2024 632 50
Voicebox: Text-Guided Multilingual Universal Speech Generation at Scale Voicebox Speech & Sound 2023 Neural Information Processing Syst 577 69
Ego-Exo4D: Understanding Skilled Human Activity from First- and Third-Person Perspectives Ego-Exo4D World Models & Embodied AI 2023 Computer Vision and Pattern Recogn 573 76
ParlAI: A Dialog Research Software Platform ParlAI Open Source Infra 2017 Conference on Empirical Methods in 405 40
Better & Faster Large Language Models via Multi-token Prediction Multi-token prediction Language & LLMs 2024 International Conference on Machin 380 44
Habitat 3.0: A Co-Habitat for Humans, Avatars and Robots Habitat World Models & Embodied AI 2023 International Conference on Learni 323 85
Emu: Enhancing Image Generation Models Using Photogenic Needles in a Haystack Emu Generative Media 2023 arXiv.org 315 12
MobileLLM: Optimizing Sub-billion Parameter Language Models for On-Device Use Cases MobileLLM On-Device & Silicon 2024 International Conference on Machin 306 20
BlenderBot 3: a deployed conversational agent that continually learns to responsibly engage BlenderBot Language & LLMs 2022 arXiv.org 300 23
Decoding speech perception from non-invasive brain recordings Brain & AI: Decoding Perception from MEG/EEG AI for Science 2022 Nature Machine Intelligence 298 38
Emu Video: Factorizing Text-to-Video Generation by Explicit Image Conditioning Emu Video & Emu Edit Generative Media 2023 European Conference on Computer Vi 296 12
Sapiens: Foundation for Human Vision Models Sapiens Vision 2024 European Conference on Computer Vi 284 44
UMA: A Family of Universal Models for Atoms UMA — Universal Models for Atoms AI for Science 2025 Neural Information Processing Syst 234 37
SAM 3D: 3Dfy Anything in Images SAM 3D (Objects + Body) Vision 2025 arXiv.org 221 50
SeamlessM4T: Massively Multilingual&Multimodal Machine Translation SeamlessM4T & the Seamless family Speech & Sound 2023 220 29
Combining Deep Reinforcement Learning and Search for Imperfect-Information Games ReBeL Games & Strategy 2020 Neural Information Processing Syst 200 16
Audiobox: Unified Audio Generation with Natural Language Prompts Audiobox Speech & Sound 2023 arXiv.org 184 11
Byte Latent Transformer: Patches Scale Better Than Tokens Byte Latent Transformer Language & LLMs 2024 arXiv.org 160 21
Open Materials 2024 (OMat24) Inorganic Materials Dataset and Models Open Materials 2024 AI for Science 2024 arXiv.org 157 12
SpiRit-LM: Interleaved Spoken and Written Language Model Spirit LM Speech & Sound 2024 Transactions of the Association fo 156 20
ELF OpenGo: An Analysis and Open Reimplementation of AlphaZero ELF OpenGo Games & Strategy 2019 International Conference on Machin 122 13
Large Concept Models: Language Modeling in a Sentence Representation Space Large Concept Models Language & LLMs 2024 arXiv.org 117 11
The Open Molecules 2025 (OMol25) Dataset, Evaluations, and Models Open Molecules 2025 AI for Science 2025 100 9
From Audio to Photoreal Embodiment: Synthesizing Humans in Conversations audio2photoreal Reality Labs Research 2024 Computer Vision and Pattern Recogn 99 14
PARTNR: A Benchmark for Planning and Reasoning in Embodied Multi-agent Tasks PARTNR World Models & Embodied AI 2024 arXiv.org 83 14
CWM: An Open-Weights LLM for Research on Code Generation with World Models Code World Model Language & LLMs 2025 arXiv.org 78 12
Meta Large Language Model Compiler: Foundation Models of Compiler Optimization Meta LLM Compiler On-Device & Silicon 2024 arXiv.org 71 2
VL-JEPA: Joint Embedding Predictive Architecture for Vision-language VL-JEPA World Models & Embodied AI 2025 arXiv.org 54 3
Video Seal: Open and Efficient Video Watermarking Video Seal Vision 2024 arXiv.org 52 12
emg2pose: A Large and Diverse Benchmark for Surface Electromyographic Hand Pose Estimation emg2pose Reality Labs Research 2024 Neural Information Processing Syst 43 4
emg2qwerty: A Large Dataset with Baselines for Touch Typing using Surface Electromyography emg2qwerty Reality Labs Research 2024 Neural Information Processing Syst 39 7
SAM Audio: Segment Anything in Audio SAM Audio Speech & Sound 2025 arXiv.org 38 3
Memory Layers at Scale Memory Layers at Scale Language & LLMs 2024 arXiv.org 35 2
Omnilingual ASR: Open-Source Multilingual Speech Recognition for 1600+ Languages Omnilingual ASR Speech & Sound 2025 arXiv.org 35 5
Brain-to-Text Decoding: A Non-invasive Approach via Typing Brain2Qwerty AI for Science 2025 arXiv.org 33 3
MobileLLM-R1: Exploring the Limits of Sub-Billion Language Model Reasoners with Open Training Recipes MobileLLM-R1 / R1.5 On-Device & Silicon 2025 arXiv.org 9 0
MobileLLM-Pro Technical Report MobileLLM-Pro On-Device & Silicon 2025 arXiv.org 4 1
Large-scale Codec Avatars: The Unreasonable Effectiveness of Large-scale Avatar Pretraining Codec Avatars Reality Labs Research 2026 arXiv.org 3 0
MobileLLM-Flash: Latency-Guided On-Device LLM Design for Industry Scale Deployment MobileLLM-Flash On-Device & Silicon 2026 Proceedings of the 64th Annual Mee 2 0

Counts move, and each is a snapshot. Citation counts favour age — a 2017 paper has had years to accumulate what a 2025 paper has had months for, so compare within a year cohort rather than across the table. Influential citations is Semantic Scholar's own measure of citations where the cited paper was central to the citing one, not merely mentioned; it is an estimate, not a count of anything a human checked. Semantic Scholar's index lags publication and misses some venues, and it occasionally holds duplicate records — one paper here reads from a second record because its arXiv entry returned zero. Only papers this catalog links are counted, so this is not all of Meta's research. Source: pulse.json.