Анатомія ШІ: дизайн концепти та процес розробки з нуля
MP3•בית הפרקים
Manage episode 480213207 series 3015412
תוכן מסופק על ידי Fwdays Tech Talks. כל תוכן הפודקאסטים כולל פרקים, גרפיקה ותיאורי פודקאסטים מועלים ומסופקים ישירות על ידי Fwdays Tech Talks או שותף פלטפורמת הפודקאסט שלהם. אם אתה מאמין שמישהו משתמש ביצירה שלך המוגנת בזכויות יוצרים ללא רשותך, אתה יכול לעקוב אחר התהליך המתואר כאן https://he.player.fm/legal.
Зустрічайте восьмий випуск Fwdays Architecture Talks! Наші постійні спікери — Олександр Савченко, Йожеф Гісем та гість випуску Дмитро Овчаренко, AI CTO of Ministry of Digital Transformation, обговорять теми: — Reference Architecture(s), Patterns, Styles для AI — Процес створення кастомної LLM - Тренування моделей - Основні quality attributes (performance, caching, availability, security, ethical aspects) — Як з'являються AI/GenAI інженери та де їх шукати? Запрошуємо вас на конференцію Highload fwdays'25: https://bit.ly/3DitOVr Корисні посилання: — AI Enterprise Architecture: - https://opea-project.github.io/latest/framework/framework.html - https://www.nvidia.com/en-us/data-center/products/ai-enterprise/ - FTI - https://learning.oreilly.com/library/view/llm-engineers-handbook/9781836200079/ — What are Large Language Models (LLMs)? by Databricks - https://www.databricks.com/glossary/large-language-models-llm — Continius pre-training vs finetuning - https://www.linkedin.com/pulse/teaching-old-dog-new-tricks-difference-between-lewis-ms-ccrp-ches-3abqc/ — Inference high-load and LLM in production - https://www.youtube.com/watch?v=NJ1jAfWR84k&list=WL&index=11&t=134s, — Recommended Book by O.Savchenko: Build a Large Language Model (From Scratch) by Sebastian Raschka - https://www.amazon.com/Build-Large-Language-Model-Scratch/dp/1633437167 - Code for developing and videos from Book - https://github.com/rasbt/LLMs-from-scratch - Author Youtube - https://www.youtube.com/@SebastianRaschka — Thoughtworks Tech Radar - AI items from ON-HOLD - https://www.thoughtworks.com/radar/techniques — ISO/IEC 42001:2023 - https://www.iso.org/ru/standard/81230.html — EU Artificial Intelligence Act - https://artificialintelligenceact.eu/ — Custom LLMs - https://er.ucu.edu.ua/server/api/core/bitstreams/1b205c1b-226c-4021-86e7-29a6063a46e6/content - GPT-NL (Netherlands) - Modello Italia (Italy, 9B parameters) - ALLaM (Saudi Arabia, 70B parameters) - SEA-Lion (Singapore, 7B parameters) - TAIDE (Taiwan, 13B parameters) - UAE’s Falcon models — Datasets info: - Malyuk Corpus: https://huggingface.co/datasets/lang-uk/malyuk - A large Ukrainian web corpus useful for pretraining foundational models. - NER-UK: https://github.com/lang-uk/ner-uk - A dataset for Named Entity Recognition (NER) covering people, organizations, and locations. - UA-GEC (Grammarly Corpus): https://github.com/grammarly/ua-gec - Focused on grammatical error correction—useful for fine-tuning language correctness models. - БрУК (Brown Ukrainian Corpus): https://github.com/brown-uk/corpus - A balanced Ukrainian text corpus for linguistic analysis and text classification tasks. - ZNO Corpus: https://huggingface.co/datasets/osyvokon/zno - Extracted from Ukrainian standardized exam texts, beneficial for educational AI applications. - Eval-UA-tion Benchmark: https://aclanthology.org/2024.unlp-1.13/ - A standardized benchmarking suite for evaluating Ukrainian LLMs across multiple NLP tasks. - UNLP-2024 Shared Task: https://github.com/unlp-workshop/unlp-2024-shared-task - A leaderboard and challenge dataset for fine-tuning Ukrainian LLMs. - Awesome Ukrainian NLP: https://github.com/osyvokon/awesome-ukrainian-nlp - A collection of Ukrainian NLP resources, pretrained models, datasets, and tools. На що варто підписатися: – Більше цікавого для розробників: https://fwdays.com – Телеграм-канал Fwdays: https://t.me/fwdays – Телеграм-канал Олексія: https://t.me/OleksiiTheArchitect – LinkedIn Олексія: https://www.linkedin.com/in/alexhelkar/ – LinkedIn Олександра: https://www.linkedin.com/in/o-savchenko/ – LinkedIn Дмитра: https://www.linkedin.com/in/dmytroovcharenko/?originalSubdomain=ua
…
continue reading
84 פרקים