Find similar content

✔️Stanford и Google представили Marin — первую полностью открытую LLM, разработанную на JAX Что делает Marin особенной: — Полностью открыты не только веса, но показан весь процесс обучения: код, данные, гиперпараметры модели, логи, эксперименты — всё доступно на GitHub — Модель обучена на 12.7 трлн токенов и в 14 из 19 тестов обошла Llama 3.1 8B — Лицензия Apache 2.0, всё можно использовать, модифицировать и воспроизводить — Levanter + JAX обеспечивают bit‑exact повторяемость и масштабируемость на TPU/GPU Проект позиционируется как открытая лаборатория: каждый эксперимент оформляется через pull request, логируется в WandB, обсуждается в issue и фиксируется в истории репозитория. Даже неудачные эксперименты сохраняются ради прозрачности. Выпущены две версии: - Marin‑8B‑Base — сильный base-модель, превосходит Llama 3.1 8B - Marin‑8B‑Instruct — обучена с помощью SFT, обгоняет OLMo 2, немного уступает Llama 3.1 Tulu Это не просто открытые веса, а новый стандарт для научных вычислений в эпоху больших моделей. * JAX — это фреймворк от Google для научных и численных вычислений, особенно популярен в сфере машинного обучения. **TPU (Tensor Processing Unit) — это специализированный чип от Google, созданный для ускорения AI-задач. 🟠Github: https://github.com/stanford-crfm/marin 🟠Блог: https://developers.googleblog.com/en/stanfords-marin-foundation-model-first-fully-open-model-developed-using-jax/ 🟠Гайд: https://docs.jax.dev/en/latest/quickstart.html @ai_machinelearning_big_data #ai#ml#tpu#jax#google

#jupyter_notebook#deep_learning#pytorch You can learn PyTorch effectively in 20 days with a friendly, well-structured guide designed for those who already know some machine learning basics and have used Keras, TensorFlow, or PyTorch before. The book breaks down PyTorch concepts from easy to hard, with clear examples and practical code you can use right away. It includes a daily plan requiring 30 minutes to 2 hours, covering modeling, core concepts, APIs, and even advanced topics like GPU training and recommendation systems. This approach makes mastering PyTorch easier and faster, helping you build strong skills for deep learning projects and real applications. https://github.com/lyhue1991/eat_pytorch_in_20_days

#вакансия#ml#rl#python#numpy#pandas#pytorch#jax#mlflow#rllib О НАШЕМ ПРОЕКТЕ Мы работаем над системой управления доходами (RMS). Наши клиенты - российские авиакомпании. Приглашаем в команду Data scientist для разработки и обучения RL-агента для управления доходами авиаперевозчика на уровне рынка. Основной стек: Python, PyTorch/JAX, NumPy/Pandas, MLFlow/Weights & Biases, приветствуются RLlib/Acme/Stable-Baselines. 🧑‍💻Чем предстоит заниматься: 🖊постановка формальной задачи RL: определение пространств состояний/действий/вознаграждения, ограничений и KPI; 🖊разработка и валидация симулятора рыночной среды на основе исторических данных (реакции спроса, сезонность, шоки); 🖊исследование и внедрение алгоритмов RL/IL (value-based, policy-gradient, actor-critic, off-policy/offline RL); 🖊экспериментальный дизайн: off-policy оценка, А/В в симуляторе, подготовка к онлайн-экспериментам; 🖊инструменты качества: стабильность обучения, воспроизводимость, мониторинг метрик (reward, RM KPI, робастность к шокам); 🖊 взаимодействие с продуктом/инженерией: требования, передача моделей в прод, контроль деградаций. 🧑‍💻Что ожидаем: 🖊сильная подготовка в RL/оптимизации/статистике (магистр/кандидат или сопоставимый опыт); 🖊практике в PyTorch/JAX; опыт построения и отладки сложных обручающих циклов; 🖊 желателен опыт causal inference/ контрафактической оценки; 🖊будет плюсом: временные ряды, эконометрика спроса, ценовые эксперименты; 🖊умение формализовать задачу и защитимо сравнивать политики. 🧑‍💻​​​​​​​Мы предлагаем: - СТАБИЛЬНОСТЬ: оформление и оклад в соответствии с ТК РФ (гпх, фриланс - невозможны); - БЕЗОПАСНОСТЬ: работа в аккредитованной IT-компании, отсрочка и т.д; - УДАЛЕННУЮ РАБОТУ: график работы 5/2 по МСК в интервале 09-18.00 -/+2 часа (гибкое начало рабочего дня с учетом планирования общих коммуникаций); - РАЗВИТИЕ: современный стек, наставничество в первый месяц работы, карьерный рост; - процессы без бюрократии, политика «открытых дверей» руководства. 📝 Ждём ваши резюме 89287653141, тг @MariP_rnd

#jupyter_notebook#mujoco#physics#robotics MuJoCo is a powerful physics engine that helps researchers and developers simulate complex movements and interactions, especially in robotics and machine learning. It provides fast and accurate simulations, which are crucial for understanding how objects move and interact with their environment. MuJoCo is beneficial because it allows users to create realistic models of multi-joint systems, compute both forward and inverse dynamics, and even handle contacts and constraints effectively. This makes it a valuable tool for those working in fields like robotics, biomechanics, and animation[1][2][5]. https://github.com/google-deepmind/mujoco

#other#automl#chatgpt#data_analysis#data_science#data_visualization#data_visualizations#deep_learning#gpt#gpt_3#jax#keras#machine_learning#ml#nlp#python#pytorch#scikit_learn#tensorflow#transformer This is a comprehensive, regularly updated list of 920 top open-source Python machine learning libraries, organized into 34 categories like frameworks, data visualization, NLP, image processing, and more. Each project is ranked by quality using GitHub and package manager metrics, helping you find the best tools for your needs. Popular libraries like TensorFlow, PyTorch, scikit-learn, and Hugging Face transformers are included, along with specialized ones for time series, reinforcement learning, and model interpretability. This resource saves you time by guiding you to high-quality, actively maintained libraries for building, optimizing, and deploying machine learning models efficiently. https://github.com/ml-tooling/best-of-ml-python

#jupyter_notebook#ai#llm#llms#multi_modal#openai#python#rag Retrieval-Augmented Generation (RAG) is a technique that helps improve the accuracy of large language models by fetching relevant information from databases or documents. This approach ensures that the model's responses are based on up-to-date and accurate data, reducing errors and "hallucinations" where the model might provide false information. For users, RAG offers more reliable and trustworthy responses, allowing them to verify the sources used to generate those responses. This method also saves resources by avoiding the need to retrain models with new data. https://github.com/FareedKhan-dev/all-rag-techniques

#python#deep_learning#diffusion#flax#flux#hacktoberfest#image_generation#image2image#image2video#jax#latent_diffusion_models#pytorch#score_based_generative_modeling#stable_diffusion#stable_diffusion_diffusers#text2image#text2video#video2video The Hugging Face Diffusers library is a powerful and easy-to-use tool for generating images, audio, and 3D molecular structures using advanced diffusion models. It offers ready-to-use pretrained models and flexible components like pipelines, schedulers, and model building blocks, allowing you to quickly create or customize your own diffusion-based projects. Installation is simple via pip or conda, and you can generate high-quality outputs with just a few lines of code. This library benefits you by making cutting-edge AI generation accessible, customizable, and efficient, whether you want to run models or train your own[1][2][5]. https://github.com/huggingface/diffusers

#jupyter_notebook#artificial_intelligence#book#large_language_models#llm#llms#oreilly#oreilly_books You can learn how to use Large Language Models (LLMs) effectively through the book *Hands-On Large Language Models* by Jay Alammar and Maarten Grootendorst. This book uses nearly 300 custom illustrations to explain key concepts and practical tools for working with LLMs, including tokenization, transformers, prompt engineering, fine-tuning, and advanced text generation. It also provides runnable code examples in Google Colab, making it easy to practice and apply what you learn. This resource helps you understand and build your own LLM applications confidently, saving you time and effort in mastering complex AI technology. It’s highly recommended for anyone wanting hands-on experience with LLMs. https://github.com/HandsOnLLM/Hands-On-Large-Language-Models

#jupyter_notebook#agentic_ai#agents#course#huggingface#langchain#llamaindex#smolagents The Hugging Face Agents Course is a free, interactive course that teaches you how to build and deploy AI agents. It's divided into four units, starting with the basics of agents and ending with a final project where you create and test your own agent. You'll learn about frameworks like `smolagents`, `LangGraph`, and `LlamaIndex`, and how to use large language models (LLMs) in your agents. The course benefits you by providing hands-on experience and practical skills in AI agent development, helping you become proficient in creating and deploying AI agents. https://github.com/huggingface/agents-course

#jupyter_notebook#aiagent#chatgpt#finance#fingpt#large_language_models#multimodal_deep_learning#prompt_engineering#robo_advisor FinRobot is a free open-source platform using AI agents and large language models for easy financial analysis. It automates stock predictions, equity reports from 10-K filings, risk checks, valuations like P/E ratios, and trading strategies with real-time data from news and markets. Install via Python, add API keys, and run demos for instant insights. This saves you hours on complex research, delivers pro-level reports and forecasts accurately, and helps make smarter investment decisions without expert skills. https://github.com/AI4Finance-Foundation/FinRobot

#jupyter_notebook#darknet#pytorch#scaled_yolov4#yolor#yolov3#yolov4#yolov7 YOLOv7 is a powerful tool for detecting objects in images and videos. It is fast, accurate, and can work well on devices with limited power, making it useful for real-time applications like self-driving cars and surveillance systems. YOLOv7 uses advanced techniques like Feature Pyramid Networks to detect objects of different sizes and can handle complex scenes with overlapping objects. This makes it beneficial for users who need quick and precise object detection in various environments. https://github.com/WongKinYiu/yolov7

#jupyter_notebook#ai#artificial_intelligence#chatgpt#deep_learning#from_scratch#gpt#language_model#large_language_models#llm#machine_learning#python#pytorch#transformer You can learn how to build your own large language model (LLM) like GPT from scratch with clear, step-by-step guidance, including coding, training, and fine-tuning, all explained with examples and diagrams. This approach mirrors how big models like ChatGPT are made but is designed to run on a regular laptop without special hardware. You also get access to code for loading pretrained models and fine-tuning them for tasks like text classification or instruction following. This helps you deeply understand how LLMs work inside and lets you create your own functional AI assistant, gaining practical skills in AI development[1][2][3][4]. https://github.com/rasbt/LLMs-from-scratch