Return to AI-DL-ML-LLM GitHub, AI-DL-ML-LLM Focused Companies, Hugging Face AI-DL-ML-LLM Services, AWS AI-DL-ML-LLM Services, Azure AI-DL-ML-LLM Services, GCP AI-DL-ML-LLM Services, IBM Cloud AI-DL-ML-LLM Services, Oracle Cloud AI-DL-ML-LLM Services, OpenAI AI-DL-ML-LLM Services, NVIDIA AI-DL-ML-LLM Services, Intel AI-DL-ML-LLM Services, Kubernetes AI-DL-ML-LLM Services, Apple AI-DL-ML-LLM Services, Meta-Facebook AI-DL-ML-LLM Services, Cisco AI-DL-ML-LLM Services
For the top 15 GitHub repos, ask for 10 paragraphs. e.g. Amazon SageMaker Features, Amazon SageMaker Alternatives, Amazon SageMaker Security, , Amazon SageMaker DevOps
In September 2024, Meta AI released Llama 3, the latest iteration of its Large Language Model (LLM) series. Llama 3 offers enhanced performance and scalability, supporting a range of applications from conversational AI to content generation. The model is available under a community license, promoting open research and development in the AI community.
Introduced in 2023, Lingua is a lean and efficient library designed for training and inference of large language models. Developed by Meta AI, Lingua utilizes modular PyTorch components, enabling researchers to experiment with new architectures and loss functions. It facilitates end-to-end training and evaluation, providing tools to enhance speed and stability in LLM research.
Launched in 2023, Pearl is an open-source library developed by Meta's Applied Reinforcement Learning team. It enables the development of reinforcement learning agents for real-world applications, emphasizing scalability and efficiency. Pearl supports various RL algorithms and provides a robust framework for deploying AI agents in production environments.
In 2024, Meta introduced MobileLLM, a toolkit aimed at optimizing sub-billion parameter language models for on-device use cases. This initiative focuses on reducing model size and computational requirements, facilitating the deployment of LLMs on mobile and edge devices without compromising performance.
Released in 2023, the Segment Anything Model (SAM) is a versatile tool for image segmentation tasks. Developed by Meta AI, SAM provides code for running inference, along with trained model checkpoints and example notebooks, enabling users to perform segmentation across diverse image datasets.
In 2023, Meta AI unveiled Chameleon, a multimodal model capable of processing and generating content across various data types, including text and images. The repository offers standalone inference code and tools for viewing multimodal inputs and outputs, facilitating research in cross-modal AI applications.
The Modular Framework (MMF), developed by Meta AI, is designed for vision and language multimodal research. MMF includes reference implementations of state-of-the-art models and has powered multiple research projects, providing a flexible platform for developing and evaluating multimodal AI systems.
Introduced in 2023, the LLM Cross Capabilities repository provides official implementations for evaluating large language models across various tasks. It includes prompts and evaluation metrics, serving as a benchmark for assessing the performance and generalization abilities of LLMs.
Building upon the original SAM, Meta AI released SAM 2 in 2024, offering improved accuracy and efficiency in image segmentation tasks. The repository provides code for running inference with SAM 2, along with trained model checkpoints and example notebooks for practical applications.
Faiss, developed by Meta AI, is a library for efficient similarity search and clustering of dense vectors. It is widely used in applications requiring fast nearest neighbor search, such as recommendation systems and image retrieval, providing scalable solutions for large datasets.
https://github.com/facebookresearch/faiss
Detectron2, introduced in 2019, is an open-source platform for object detection and segmentation. It supports a range of state-of-the-art models and is widely used in Computer Vision research and applications.
PyTorch3D, launched in 2020, is a library for 3D deep learning and rendering. It provides tools for working with 3D data, enabling tasks like shape reconstruction, pose estimation, and scene understanding.
Fairseq, introduced in 2017, is a sequence-to-sequence learning toolkit for Natural Language Processing (NLP). It supports Transformer models and various tasks, including translation, summarization, and text generation.
LASER, launched in 2019, is a toolkit for generating multilingual sentence embeddings. It supports over 90 languages and enables cross-lingual applications like translation and multilingual search.
PyText, introduced in 2018, is a Natural Language Processing (NLP) framework built on PyTorch. It simplifies the development and deployment of custom NLP models, including classifiers and language understanding systems.
DINO, launched in 2021, applies self-supervised learning to Vision Transformers (ViT). It enables effective feature extraction without requiring labeled data, advancing research in unsupervised Computer Vision.
BlenderBot, introduced in 2020, is an open-domain chatbot built using Large Language Models (LLMs). It focuses on generating engaging and contextually relevant responses in conversational AI systems.
https://github.com/facebookresearch/ParlAI/tree/master/projects/recipes/blenderbot
XLM-R, launched in 2019, is a cross-lingual language model supporting over 100 languages. It is optimized for tasks like multilingual understanding, translation, and question answering.
Hydra, introduced in 2020, is a framework for building and managing complex configurations in Machine Learning (ML). It simplifies experimentation by supporting dynamic configurations for models, data, and parameters.
MemN2N, launched in 2015, provides tools for training memory networks for reasoning tasks. It is widely used in question answering and knowledge base applications, showcasing advancements in AI reasoning.
https://github.com/facebookresearch/memnn
DeepFocus, introduced in 2019, is a computational imaging system for virtual reality. It uses AI and Deep Learning (DL) techniques to create advanced focus cues for improving immersion in virtual reality environments.
VisSL, launched in 2021, is a framework for self-supervised learning on visual data. It supports large-scale pretraining for Computer Vision tasks, providing tools for experimentation with various self-supervised methods.
FastText, introduced in 2016, is an efficient library for text classification and word representation. It is widely used in Natural Language Processing (NLP) tasks like sentiment analysis, language detection, and text similarity.
N-BEATS, launched in 2020, is a deep neural architecture for time-series forecasting. It is designed for accuracy and scalability, supporting tasks like demand prediction, financial forecasting, and weather modeling.
Open Catalyst Project, introduced in 2020, is a collaborative effort to use AI for discovering catalysts for renewable energy solutions. It provides large datasets and models for simulating catalytic processes.
ParlAI, launched in 2017, is a framework for dialogue research. It supports training and evaluating conversational agents, integrating with tasks like dialogue modeling, question answering, and chatbot development.
WaveRNN, introduced in 2018, is a lightweight Text-to-Speech (TTS) model. It provides high-quality audio synthesis with reduced computational requirements, enabling deployment on mobile and edge devices.
Acoustic Modeling Toolkit (AMT), launched in 2020, is a library for training AI models in speech processing. It supports tasks like speech recognition, noise suppression, and voice activity detection.
Dynabench, introduced in 2021, is a platform for benchmarking AI models in dynamic, real-world scenarios. It supports interactive testing for tasks like natural language inference, sentiment analysis, and image classification.
FAIR Scale, launched in 2020, is a toolkit for large-scale distributed training of deep learning models. It integrates with PyTorch and includes optimizations for scaling to thousands of GPUs in research and production environments.
https://github.com/facebookresearch/fairscale
PyTorch-BigGraph, introduced in 2019, is a framework for training embeddings of large graphs. It is optimized for handling billions of nodes and edges, supporting applications like knowledge graph embedding and link prediction.
FBLearner Flow, launched in 2016, is Meta's end-to-end machine learning platform. It provides tools for feature engineering, model training, and deployment, streamlining the entire machine learning lifecycle.
Ax, introduced in 2018, is a platform for managing adaptive experiments. It supports optimization tasks like hyperparameter tuning and A/B testing, helping researchers and practitioners improve model performance.
PyTorch Elastic, launched in 2020, provides tools for dynamic scaling of distributed training jobs. It enables fault-tolerant training workflows for Deep Learning (DL) models on cloud and on-premises environments.
HydraZen, introduced in 2021, extends Hydra for easier and safer management of machine learning configurations. It provides a user-friendly interface for handling complex model and experiment configurations.
Adversarial Robustness Toolbox (ART), launched in 2018, offers tools for evaluating and improving the robustness of AI models against adversarial attacks. It supports tasks like image classification and speech recognition.
https://github.com/facebookresearch/Adversarial-Robustness-Toolbox
Video Understanding Framework, introduced in 2020, is a library for building and evaluating models for video analysis. It supports tasks like action recognition, temporal segmentation, and video captioning.
PySlowFast, launched in 2020, is a video understanding codebase based on the SlowFast networks. It enables efficient training and evaluation of models for tasks like video classification and activity recognition.
ELF, introduced in 2018, is a research platform for game AI and reinforcement learning. It supports tasks like strategy evaluation and learning in complex multi-agent environments.
BoTorch, launched in 2019, is a library for Bayesian optimization built on PyTorch. It supports tasks like hyperparameter tuning and experimental design, leveraging advanced probabilistic models.
https://github.com/facebook/botorch
Animated Drawings, introduced in 2022, is a toolkit for turning human-drawn characters into animated figures. It uses Computer Vision techniques to segment and animate 2D images, enabling creative applications in art and storytelling.
Commerce Detection Toolkit, launched in 2021, provides tools for detecting product-related entities in text and images. It supports e-commerce applications like catalog generation, product recommendations, and automated tagging.
Adversarial NLI (ANLI), introduced in 2020, is a dataset and benchmark for evaluating robustness in Natural Language Understanding (NLU). It focuses on testing the generalization of AI models under adversarial scenarios.
Ego4D Dataset, launched in 2021, is a large-scale egocentric video dataset for training AI models in first-person perception tasks. It supports applications like activity recognition, object tracking, and video summarization.
R2D3, introduced in 2020, is a reinforcement learning framework tailored for text-based games. It combines textual input processing with action generation, enabling models to learn decision-making in narrative environments.
Opacus, launched in 2020, is a library for training machine learning models with differential privacy. It integrates seamlessly with PyTorch and supports tasks requiring strict privacy guarantees.
Roberta, introduced in 2019, is a robustly optimized variant of BERT for pretraining NLP models. It improves performance on a wide range of benchmarks by leveraging larger datasets and better training techniques.
https://github.com/pytorch/fairseq/tree/master/examples/roberta
LightConv and DynamicConv, launched in 2019, are lightweight convolutional models for sequence modeling. They provide efficient alternatives to Transformer architectures for tasks like machine translation and text summarization.
https://github.com/pytorch/fairseq/tree/master/examples/lightconv_dynamicconv
DeepMask and SharpMask, introduced in 2016, are models for instance segmentation in Computer Vision. They provide pixel-level masks for objects in images, advancing research in object detection and recognition.
Demucs, launched in 2019, is a deep learning-based audio source separation model. It is designed to isolate vocals, drums, and other instruments from music tracks, supporting applications in music production and analysis.
https://github.com/facebookresearch/demucs
Fairseq S2T, introduced in 2020, is an extension of Fairseq for speech-to-text processing. It supports tasks like speech recognition and translation, providing models and tools for end-to-end training of audio-based systems.
https://github.com/facebookresearch/fairseq/tree/master/examples/speech_to_text
Deep Learning Recommendation Model (DLRM), launched in 2019, is a toolkit for building large-scale recommendation systems. It integrates with PyTorch and supports scalable training for personalized content delivery.
Detectron, introduced in 2018, is an earlier object detection platform that supports models like Faster R-CNN and Mask R-CNN. It served as the foundation for the development of Detectron2, advancing research in Computer Vision.
PyRobot, launched in 2019, is a Python-based interface for robotics. It simplifies access to robotics platforms, enabling quick prototyping of AI algorithms in robotic perception, control, and planning.
WizLang, introduced in 2021, is a toolkit for experimenting with few-shot learning in NLP. It provides benchmarks and datasets to evaluate the ability of language models to generalize from limited examples.
Habitat, launched in 2019, is a platform for training embodied AI agents in 3D virtual environments. It supports tasks like navigation, object manipulation, and semantic understanding in realistic indoor simulations.
Fairsim, introduced in 2020, is a simulation platform for large-scale distributed reinforcement learning. It supports high-throughput experimentation for multi-agent and single-agent AI systems in dynamic environments.
Loop, launched in 2022, is a framework for continuous integration and evaluation of AI models. It ensures consistent performance through automated retraining and validation workflows, reducing deployment friction.
ReAgent, introduced in 2019, is an open-source platform for reinforcement learning-based applications. It integrates with PyTorch and supports tasks like personalized recommendations, dynamic pricing, and adaptive systems.
XLM, launched in 2019, is a language model designed for cross-lingual tasks. It supports applications in multilingual machine translation and language understanding, enabling seamless transfer between languages.
https://github.com/facebookresearch/XLM
TorchServe, introduced in 2020, is an open-source model serving framework for deploying PyTorch models at scale. It supports high-performance inference, multi-model serving, and model management for production environments.
Bean Machine, launched in 2021, is a probabilistic programming framework for Bayesian machine learning. It simplifies the creation and training of probabilistic models, supporting applications like causal inference and uncertainty quantification.
Fairmotion, introduced in 2020, is a Python library for motion capture and animation analysis. It is designed for tasks like motion synthesis, retargeting, and evaluation in applications involving human and character animation.
Neural Sparse Training, launched in 2021, is a toolkit for experimenting with sparsity in Deep Learning (DL) models. It focuses on reducing model size and computational overhead while maintaining performance in various AI tasks.
Dynabench Tasks, introduced in 2020, provides resources for dynamic benchmarking of AI models across evolving tasks. It supports interactive evaluation for Natural Language Processing (NLP), vision, and multimodal AI systems.
Horizon, launched in 2018, is an open-source platform for applied reinforcement learning at scale. It is designed for real-world applications in areas like recommendations, search, and personalization, integrating seamlessly with production systems.
Aroma, introduced in 2019, is a tool for retrieving relevant code snippets based on task descriptions. It aids developers in navigating large codebases by leveraging AI for semantic code understanding and retrieval.
Video Modeling Toolkit, launched in 2021, is a framework for building generative and predictive models for video data. It supports tasks like frame synthesis, motion prediction, and video anomaly detection.
KILT, introduced in 2020, is a benchmark and framework for training and evaluating AI models on tasks requiring external knowledge. It integrates with datasets and tasks for retrieval, reasoning, and generation.