Локальные ИИ-агенты - ВСЕ, что тебе нужно знать - параметры, железо, скорость, приватность

Will local AI replace ChatGPT and Claude subscriptions? With the boom in open-source models and falling hardware requirements, the idea of ​​running a powerful neural network directly on your computer is becoming increasingly feasible. This is a complete guide to running local LLM on your PC or Mac. We'll cover 5-bit inference, VRAM, quantization, GGUF and MLX formats, and compare all the current hardware required for running it: Nvidia (RTX 3090, 4090), AMD, Intel, and shared-memory systems like Mac Studio. We'll also consider renting cloud GPUs as an alternative to purchasing them. Finally, we'll test different models (Qwen, Ornis, Gemma) on a real-world task to understand the level of intelligence you can achieve on your device. Rent a server with a graphics card to run AI models: https://slc.tl/a2zl6 Deploy a ready-made AI model in the cloud: https://slc.tl/yx8qd Advertising. JSC Selectel, Taxpayer Identification Number (INN) 7810962785. erid: CQH36pWzJqV7DdPpGnrik91w6eJTS96AbQLzRzFLu5ApS9 ⚡️ Intensive AI Development Course. Learn how to create, deploy, and monetize AI products from scratch - https://t.me/how2ai_bot?start=dl-1783... #ai #aiagent #claude Timecodes: 00:00 - Intro 01:32 - 5 basic terms: inference, VRAM, quantization... 08:44 - What hardware do you need for local AI? 15:05 - Renting cloud GPUs: When is it profitable? 16:31 - Model tests: What can local LLMs really do? 18:44 - GPT-OSS 120B Test 22:32 - GLM-5.2 Test 23:39 - Qwen3.5-4b 24:44 - Ornith 1.0 9b 25:33 - Gemma 4 12b and 26b 28:03 - Qwen 3.6 27b 29:34 - Qwen 3.6 35b a3b 30:20 - Ornith 1.0 35b 32:27 - Conclusions: Will Local AI Replace Subscriptions? Boosty – https://boosty.to/prodadvice/single-p... How2AI Telegram – https://t.me/How2AI How2AI Chat Telegram – https://t.me/how_to_AI Creative Advice Telegram – https://t.me/creadvice 🥸 Uncle D's Personal Telegram – https://t.me/true_dyadya_d