machine-learning
A tiny scalar-valued autograd engine and a neural net library on top of it with PyTorch-like API.
A tiny Autograd engine (with a bite! :)). Implements backpropagation (reverse-mode autodiff) over a dynamically built DAG and a small neural networks library on top of it with a PyTorch-like API. Both are tiny, with about 100 and 50 lines of code respectively. The DAG only operates over scalar values, so e.g. we chop up each neuron into all of its individual tiny adds and multiplies. However, this is enough to build up entire deep neural nets doing binary classification, as the demo notebook shows. Potentially useful for educational purposes.
Related contents:
A Flexible Framework for Experiencing Cutting-edge LLM Inference Optimizations.
KTransformers, pronounced as Quick Transformers, is designed to enhance your 🤗 Transformers experience with advanced kernel optimizations and placement/parallelism strategies.
KTransformers is a flexible, Python-centric framework designed with extensibility at its core. By implementing and injecting an optimized module with a single line of code, users gain access to a Transformers-compatible interface, RESTful APIs compliant with OpenAI and Ollama, and even a simplified ChatGPT-like web UI.
Artificial Neural Engine Machine Learning Library.
ANEMLL (pronounced like "animal") is an open-source project focused on accelerating the porting of Large Language Models (LLMs) to tensor processors, starting with the Apple Neural Engine (ANE).
Look At Your Data 👀.
Data quality is the most important factor in machine learning success. Hyperparam brings exploration and analysis of massive text datasets to the browser.
I made my AI think harder by making it argue with itself repeatedly. It works stupidly well.
CoRT makes AI models recursively think about their responses, generate alternatives, and pick the best one. It's like giving the AI the ability to doubt itself and try again... and again... and again.
A machine learning framework for Node.js, based on MLX.
A self-contained, lightweight and OOB research platform for modern ML.
Boson is a lightweight, fully containerized, and feature-rich machine learning research platform. It centralizes essential tools to help teams keep projects lean, organized, and reproducible—while reducing overhead and boosting productivity. Think Databricks/Sagemaker but local and free.
Boson enables engineers and researchers to iterate faster without getting bogged down by infrastructure or tooling complexity.
ML Pipelines From Another Planet.Build out-of-this-world ML pipelines.
Run-anywhere computational framework for Python that simplifies and accelerates ML workflows and development. xorq is a deferred computational framework for building, running, and serving pandas groupby-apply style pipelines common in ML workflows. xorq is built on top of Ibis and Apache DataFusion.
JobSet: a k8s native API for distributed ML training and HPC workloads
JobSet is a Kubernetes-native API for managing a group of k8s Jobs as a unit. It aims to offer a unified API for deploying HPC (e.g., MPI) and AI/ML training workloads (PyTorch, Jax, Tensorflow etc.) on Kubernetes.
Related contents:
A Datacenter Scale Distributed Inference Serving Framework.
NVIDIA Dynamo is a high-throughput low-latency inference framework designed for serving generative AI and reasoning models in multi-node distributed environments. Dynamo is designed to be inference engine agnostic (supports TRT-LLM, vLLM, SGLang or others) and captures LLM-specific capabilities.
Related contents:
A Friendly Federated AI Framework.
A unified approach to federated learning, analytics, and evaluation. Federate any workload, any ML framework, and any programming language.
Free & Open-Source AI Voice Generator.
A powerful, browser-based AI voice generator that lets you create natural-sounding voices without installing anything.
Use it directly in your browser or self-host it for your own applications with OpenAI API compatibility!
WAGMIOS is a self-hosted container management system with AI-powered automation. It enables you to efficiently manage your containers with W.I.L.L.O.W, an AI assistant that optimizes your workflow.
Evolving agents is a production-grade environment for orchestrating, evolving, and managing AI agents.
A production-grade framework for creating, managing, and evolving AI agents with intelligent agent-to-agent communication. The framework enables you to build collaborative agent ecosystems that can semantically understand requirements, evolve based on past experiences, and communicate effectively to solve complex tasks.
The Platform for Building Stateful Agents. Build agents with infinite context and human-like memory, that can learn from data and improve with experience. Letta (formerly MemGPT) is a framework for creating LLM services with memory.
👾 Letta is an open source framework for building stateful LLM applications. You can use Letta to build stateful agents with advanced reasoning capabilities and transparent long-term memory. The Letta framework is white box and model-agnostic.
Related contents:
Finding the Scaling Laws of Agents. The first and the best multi-agent framework.
🐫 CAMEL is an open-source community dedicated to finding the scaling laws of agents. We believe that studying these agents on a large scale offers valuable insights into their behaviors, capabilities, and potential risks. To facilitate research in this field, we implement and support various types of agents, tasks, prompts, models, and simulated environments.
The framework enables multi-agent systems to continuously evolve by generating data and interacting with environments. This evolution can be driven by reinforcement learning with verifiable rewards or supervised learning.
superglue is an open-source server that sits as a layer between complex APIs and your application. With superglue, you always get the data that you want in the format that you expect. Fetch data from JSON and XML APIs, as well as CSV and Excel files in seconds.
The Open-Source LLM Evaluation Framework.
DeepEval is a simple-to-use, open-source LLM evaluation framework, for evaluating and testing large-language model systems. It is similar to Pytest but specialized for unit testing LLM outputs. DeepEval incorporates the latest research to evaluate LLM outputs based on metrics such as G-Eval, hallucination, answer relevancy, RAGAS, etc., which uses LLMs and various other NLP models that runs locally on your machine for evaluation.
The TypeScript framework for agents & workflows with react-like components. Lightning fast dev loop. Easy to learn. Easy to extend.
Build complex AI applications with React-like components. GenSX is a simple typescript framework for building agents and workflows with reusable React-like components. GenSX takes a lot of inspiration from React, but the programming model is very different - it’s a Node.js framework designed for data flow.
Zonos-v0.1 is a leading open-weight text-to-speech model trained on more than 200k hours of varied multilingual speech, delivering expressiveness and quality on par with—or even surpassing—top TTS providers.
Our model enables highly natural speech generation from text prompts when given a speaker embedding or audio prefix, and can accurately perform speech cloning when given a reference clip spanning just a few seconds. The conditioning setup also allows for fine control over speaking rate, pitch variation, audio quality, and emotions such as happiness, fear, sadness, and anger. The model outputs speech natively at 44kHz.
Related contents:
BirdNET-Analyzer is an open source tool for analyzing bird calls using machine learning models. It can process large amounts of audio recordings and identify (bird) species based on their calls.
The fast, Open Source and easy-to-use solver. Solve planning and scheduling problems with OptaPlanner.
A fast, easy-to-use, open source AI constraint solver for software developers
Related contents:
build ml models in natural language and minimal code.
Create machine learning models with minimal code by describing what you want them to do in plain words. You explain the task, and the library builds a model for you, including data generation, feature engineering, training, and packaging.
For better or for worse, LLMs are here to stay. We all read content that they produce online, most of us interact with LLM chatbots, and many of us use them to produce content of our own.
In a series of five- to ten-minute lessons, we will explain what these machines are, how they work, and how to thrive in a world where they are everywhere.
You will learn when these systems can save you a lot of time and effort. You will learn when they are likely to steer you wrong. And you will discover how to see through the hype to tell the difference. ?
AI by Hand ✍️ Exercises in Excel
A Systems View of LLMs on TPUs.
This book aims to demystify the art of scaling LLMs on TPUs. We try to explain how TPUs work, how LLMs actually run at scale, and how to pick parallelism schemes during training and inference that avoid communication bottlenecks.
Open Universal Machine Intellingence. E2E Foundation Model Research Platform. Everything you need to build state-of-the-art foundation models, end-to-end.
Oumi is a fully open-source platform that streamlines the entire lifecycle of foundation models - from data preparation and training to evaluation and deployment. Whether you're developing on a laptop, launching large scale experiments on a cluster, or deploying models in production, Oumi provides the tools and workflows you need.
State-of-the-art Machine Learning for PyTorch, TensorFlow, and JAX.
🤗 Transformers provides APIs and tools to easily download and train state-of-the-art pretrained models. Using pretrained models can reduce your compute costs, carbon footprint, and save you the time and resources required to train a model from scratch.
Related contents:
ONNX Runtime: cross-platform, high performance ML inferencing and training accelerator.
ONNX Runtime inference can enable faster customer experiences and lower costs, supporting models from deep learning frameworks such as PyTorch and TensorFlow/Keras as well as classical machine learning libraries such as scikit-learn, LightGBM, XGBoost, etc.
Related contents:
Easy, fast, and cheap LLM serving for everyone.
vLLM is a fast and easy-to-use library for LLM inference and serving.
Originally developed in the Sky Computing Lab at UC Berkeley, vLLM has evloved into a community-driven project with contributions from both academia and industry.
Related contents:
- How to serve LLMs with vLLM and OVHcloud AI Deploy @ OVHcloud.
- Episode 616: From Boston to bootc @ Linux Unplugged.
- What is vLLM @ RedHat.
- Faire tourner un LLM localement sur votre ordinateur @ Quoi de neuf les devs ? :fr:.
- Inside vLLM: Anatomy of a High-Throughput LLM Inference System @ Aleksa Gordić blog.
- vLLM : Maîtriser l'Inference Haute Performance pour les LLM @ DevSecOps :fr:.
- Docker Model Runner now supports vLLM on Windows @ Docker.
- Running a self-hosted LLM in Kubernetes with vLLM @ CNCF.
Open Repository of Web Crawl Data.
Common Crawl maintains a free, open repository of web crawl data that can be used by anyone.
Related contents:
15 trillion tokens of the finest data the 🌐 web has to offer.
The 🍷 FineWeb dataset consists of more than 15T tokens of cleaned and deduplicated english web data from CommonCrawl. The data processing pipeline is optimized for LLM performance and ran on the 🏭 datatrove library, our large scale data processing library.
🍷 FineWeb was originally meant to be a fully open replication of 🦅 RefinedWeb, with a release of the full dataset under the ODC-By 1.0 license. However, by carefully adding additional filtering steps, we managed to push the performance of 🍷 FineWeb well above that of the original 🦅 RefinedWeb, and models trained on our dataset also outperform models trained on other commonly used high quality web datasets (like C4, Dolma-v1.6, The Pile, SlimPajama, RedPajam2) on our aggregate group of benchmark tasks.
Related contents:
Partner of Accounting Leaders. Generative AI platform for intelligent accounting.
The preferred partner of accounting leaders.
Related contents:
The Annual Conference on Neural Information Processing Systems.
Related content:
Run AI with an API. Run and fine-tune open-source models. Deploy custom models at scale. All with one line of code.
Thousands of models contributed by our community. All the latest open-source models are on Replicate. They’re not just demos — they all actually work and have production-ready APIs.
AI shouldn’t be locked up inside academic papers and demos. Make it real by pushing it to Replicate.
Related contents:
Discover, download, and run local LLMs.
Related contents:
- #104 Développer des projets IA - introduction @ Double Slash :fr:.
- Faire tourner un LLM localement sur votre ordinateur @ Quoi de neuf les devs ? :fr:.
- Drames et dramas d’août @ Le RDV Tech podcast.
- LM Studio : Faire tourner son IA (LLM) facilement (Chat, Developpement, ...) @ Adrien Linuxtricks' YouTube :fr:.
A fast tool to read text-based files in a repository or directory, chunk them, and serialize them for LLM consumption.
Explore Agent Recipes
Explore common agent recipes with ready to copy code to improve your LLM applications.
Related contents:
Stable Point Aware 3D (SPAR3D) can make real-time edits and create the complete structure of a 3D object from a single image in a few seconds. SPAR3D combines the strengths of point-cloud diffusion (probabilistic) and mesh regression (deterministic) to have improved details on the unseen back regions in the input image.
Related contents:
structured-logprobs is an open-source Python library that enhances OpenAI's structured outputs by providing detailed information about token log probabilities.
This library is designed to offer valuable insights into the reliability of an LLM's structured outputs. It works with OpenAI's Structured Outputs, a feature that ensures the model consistently generates responses adhering to a supplied JSON Schema. This eliminates concerns about missing required keys or hallucinating invalid values.
We wrote this glossary to solve a problem we ran into working with GPUs here at Modal : the documentation is fragmented, making it difficult to connect concepts at different levels of the stack, like Streaming Multiprocessor Architecture , Compute Capability , and nvcc compiler flags .
Open weights LLM for French, English, German, Spanish and Italian.
Related contents:
LLM génératifs ouverts et performants.
Le Consortium OpenLLM France réunit 17 acteurs qui se sont rassemblés dans le prolongement de la création de la communauté OpenLLM France qui fédère à ce jour un écosystème de près de 200 entités (laboratoires publics de recherche, fournisseurs potentiels de données, acteurs technologiques spécialisés, fournisseurs de cas d'usage...). Ces acteurs échangent de manière publique et transparente depuis le début de l’été 2023 sur le serveur Discord de la communauté.
🐢 Open-Source Evaluation & Testing for AI & LLM systems.
The testing framework dedicated to ML models, from tabular to LLMs. Control risks of performance, bias and security issues in AI systems.
High performance array computing.
Composable transformations of Python+NumPy programs: differentiate, vectorize, JIT to GPU/TPU, and more
Related contents:
Fetch an entire site and save it as a text file (to be used with AI models).
MLOps Attack Toolkit
MLOKit is a toolkit that can be used to attack MLOps platforms by taking advantage of the available REST API. This tool allows the user to specify an attack module, along with specifying valid credentials (API key or stolen access token) for the respective MLOps platform. The attack modules supported include reconnaissance, data extraction and model extraction. MLOKit was built in a modular approach, so that new modules can be added in the future by the information security community.
Take control of the media you consume every day with Fast Music Remover!.
A C++ based, lightweight music and noise remover for YouTube and other internet media, using DeepFilterNet for audio enhancement.
We consume, willingly or not, large amounts of media everyday, and that includes content that is emposed on us. Fast Music Remover gives you the choice to opt-out of them without missing out on the core content.
We're building a feature rich media processor that is efficient, modular and cross platform. It's being built for you! That means clean APIs for programmers, containerized on GHCR for remote users, with a Web UI providing seamless access to anyone interested!
NVIDIA Cosmos™ is a platform of state-of-the-art generative world foundation models, advanced tokenizers, guardrails, and an accelerated data processing and curation pipeline for autonomous vehicles (AVs) and robotics developers.
Cosmos is a world model development platform that consists of world foundation models, tokenizers and video processing pipeline to accelerate the development of Physical AI at Robotics & AV labs. Cosmos is purpose built for physical AI. The Cosmos repository will enable end users to run the Cosmos models, run inference scripts and generate videos.
Related contents:
A course on neural networks that starts all the way at the basics. The course is a series of YouTube videos where we code and train neural networks together. The Jupyter notebooks we build in the videos are then captured here inside the lectures directory. Every lecture also has a set of exercises included in the video description. (This may grow into something more respectable).
MMAudio generates synchronized audio given video and/or text inputs.
Taming Multimodal Joint Training for High-Quality Video-to-Audio Synthesis.
Related contents:
Euclid scientists need your help. Euclid captures images of tens of millions of galaxies like those we’re showing here. To classify that impossibly-large pile of galaxies, we’re using your classifications to train AI algorithms (remember the Zoobot AI assistant?). But the AI algorithms need to be ready for the scientists by the end of August - in only one month! We need as many volunteer classifications for teaching the AI algorithms as we can get; our goal is 100,000 classifications. We’re calling this The Euclid Challenge. Spread the word and dive in!
OpenVINO™ is an open-source toolkit for optimizing and deploying AI inference.
OpenVINO is an open-source toolkit for optimizing and deploying deep learning models from cloud to edge. It accelerates deep learning inference across various use cases, such as generative AI, video, audio, and language with models from popular frameworks like PyTorch, TensorFlow, ONNX, and more. Convert and optimize models, and deploy across a mix of Intel® hardware and environments, on-premises and on-device, in the browser or in the cloud.
Multilingual Multi-corpus Speech Emotion Recognition Toolkit and Benchmark.
EmoBox, a groundbreaking multilingual multi-corpus speech emotion recognition (SER) toolkit designed to streamline research in this field. EmoBox is accompanied by a meticulously curated benchmark tailored for both intra-corpus and cross-corpus evaluation settings.
QRNet is a state-of-the-art neural network that transforms the Quad-Bayer pattern into sharp and clean RGB images
A smarter web fuzzing tool that combines local LLM models and ffuf to optimize directory and file discovery.
This tool enhances traditional web fuzzing by using local AI language models (via Ollama) to generate intelligent guesses for potential paths and filenames.