Telegram Group & Telegram Channel
🌟MiniMax-M1: открытя reasoning‑LLM с контекстом 1M

MiniMax-M1 — первая в мире open-weight гибридная reasoning‑LLM c 1M контекстом (8× DeepSeek R1) и гибридной архитектурой MoE + lightning attention.
• 456 млрд параметров (45,9 млрд активируются на токен), сверхэффективная генерация — 25% FLOPs DeepSeek R1 на 100K токенов
• Обучение через RL с новым алгоритмом CISPO, решающим реальные задачи от математики до кодинга
• На обучение было потрачено $534K, две версии — 40K/80K “thinking budget”
• Обходит DeepSeek R1 и Qwen3-235B на бенчмарках по математике и кодингу,
• Топ результат на задачах для software engineering и reasoning



Бенчмарки:
AIME 2024: 86.0 (M1-80K) vs 85.7 (Qwen3) vs 79.8 (DeepSeek R1)

SWE-bench Verified: 56.0 vs 34.4 (Qwen3)

OpenAI-MRCR (128k): 73.4 vs 27.7 (Qwen3)

TAU-bench (airline): 62.0 vs 34.7 (Qwen3)

LongBench-v2: 61.5 vs 50.1 (Qwen3)


➡️ Попробовать можно здесь

Hugging Face: https://huggingface.co/collections/MiniMaxAI/minimax-m1-68502ad9634ec0eeac8cf094
GitHub: https://github.com/MiniMax-AI/MiniMax-M1
Tech Report: https://github.com/MiniMax-AI/MiniMax-M1/blob/main/MiniMax_M1_tech_report.pdf


@ai_machinelearning_big_data

#llm #reasoningmodels #minimaxm1
Please open Telegram to view this post
VIEW IN TELEGRAM
Please open Telegram to view this post
VIEW IN TELEGRAM



tg-me.com/ai_machinelearning_big_data/7799
Create:
Last Update:

🌟MiniMax-M1: открытя reasoning‑LLM с контекстом 1M

MiniMax-M1 — первая в мире open-weight гибридная reasoning‑LLM c 1M контекстом (8× DeepSeek R1) и гибридной архитектурой MoE + lightning attention.
• 456 млрд параметров (45,9 млрд активируются на токен), сверхэффективная генерация — 25% FLOPs DeepSeek R1 на 100K токенов
• Обучение через RL с новым алгоритмом CISPO, решающим реальные задачи от математики до кодинга
• На обучение было потрачено $534K, две версии — 40K/80K “thinking budget”
• Обходит DeepSeek R1 и Qwen3-235B на бенчмарках по математике и кодингу,
• Топ результат на задачах для software engineering и reasoning



Бенчмарки:
AIME 2024: 86.0 (M1-80K) vs 85.7 (Qwen3) vs 79.8 (DeepSeek R1)

SWE-bench Verified: 56.0 vs 34.4 (Qwen3)

OpenAI-MRCR (128k): 73.4 vs 27.7 (Qwen3)

TAU-bench (airline): 62.0 vs 34.7 (Qwen3)

LongBench-v2: 61.5 vs 50.1 (Qwen3)


➡️ Попробовать можно здесь

Hugging Face: https://huggingface.co/collections/MiniMaxAI/minimax-m1-68502ad9634ec0eeac8cf094
GitHub: https://github.com/MiniMax-AI/MiniMax-M1
Tech Report: https://github.com/MiniMax-AI/MiniMax-M1/blob/main/MiniMax_M1_tech_report.pdf


@ai_machinelearning_big_data

#llm #reasoningmodels #minimaxm1

BY Machinelearning





Share with your friend now:
tg-me.com/ai_machinelearning_big_data/7799

View MORE
Open in Telegram


Machinelearning Telegram | DID YOU KNOW?

Date: |

Export WhatsApp stickers to Telegram on iPhone

You can’t. What you can do, though, is use WhatsApp’s and Telegram’s web platforms to transfer stickers. It’s easy, but might take a while.Open WhatsApp in your browser, find a sticker you like in a chat, and right-click on it to save it as an image. The file won’t be a picture, though—it’s a webpage and will have a .webp extension. Don’t be scared, this is the way. Repeat this step to save as many stickers as you want.Then, open Telegram in your browser and go into your Saved messages chat. Just as you’d share a file with a friend, click the Share file button on the bottom left of the chat window (it looks like a dog-eared paper), and select the .webp files you downloaded. Click Open and you’ll see your stickers in your Saved messages chat. This is now your sticker depository. To use them, forward them as you would a message from one chat to the other: by clicking or long-pressing on the sticker, and then choosing Forward.

Look for Channels Online

You guessed it – the internet is your friend. A good place to start looking for Telegram channels is Reddit. This is one of the biggest sites on the internet, with millions of communities, including those from Telegram.Then, you can search one of the many dedicated websites for Telegram channel searching. One of them is telegram-group.com. This website has many categories and a really simple user interface. Another great site is telegram channels.me. It has even more channels than the previous one, and an even better user experience.These are just some of the many available websites. You can look them up online if you’re not satisfied with these two. All of these sites list only public channels. If you want to join a private channel, you’ll have to ask one of its members to invite you.

Machinelearning from hk


Telegram Machinelearning
FROM USA