Wait nvm, it's just a framework lol
Orange Site:
https://news.ycombinator.com/item?id=42922989
AI made in 🇪🇺
— European Commission (@EU_Commission) February 3, 2025
OpenEuroLLM, the first family of open source Large Language Models covering all EU languages, has earned the first STEP Seal for its excellence.
It brings together EU startups, research labs and supercomputing hosts to train AI on European supercomputers ↓ pic.twitter.com/9YvWBW1CpL
https://boards.4chan.org/g/thread/104200209
https://lemmy.world/post/25074353
Jump in the discussion.
No email address required.
Jump in the discussion.
No email address required.
Didn't the Chinese do it for $6 million?
Jump in the discussion.
No email address required.
The pre-training run for that specific model they published (r1) would have cost $6M if they had done it on EC2 or something similar.
But that wasn't their first attempt, it probably took 200 tries (starting smaller of course), so the (EC2 equivalent) cost of figuring out how to eventually train their r1 model was another $250M, and in order to iterate through those attempts quickly they needed a lot of computing resources, worth around $2B (which they will continue to use for the next 5 years).
That's the smallest order of magnitude that currently has a chance, but going forward it won't be enough.
OpenAI's computing resources are closer to $75B (increasing rapidly). OpenAI will quickly figure out how deepseek did it, and apply those modifications to their own upcoming models, but with 40 times as much compute as deepseek has used, so they can iterate 40 times faster at the same size as deepsekk, and eventually go much bigger than r1 (and perhaps at the end distill down to a terabyte sized model, for cheaper inference).
(Not just for openai but all its big competitors.)
Jump in the discussion.
No email address required.
Accurate. Buy the NVDA dip or get dusted.
Jump in the discussion.
No email address required.
More options
Context
More options
Context
can the yuroes do it for $37 mil?
Jump in the discussion.
No email address required.
Sure, just download DeepSeek and run it, plus $37 million to verify regulatory compliance.
Jump in the discussion.
No email address required.
Jump in the discussion.
No email address required.
More options
Context
More options
Context
More options
Context
More options
Context
Is that for the building rent to stick the LLM datacenter inside?
Jump in the discussion.
No email address required.
More options
Context
More options
Context