mag retweeted
Whoever convinced people that Huggingface is named after an emoji instead of the creature from Alien is the greatest liar uh I mean the greatest marketer of all time
mag retweeted
Massive power blackout in Spain and Portugal! No internet, payment are off and flights are cancelled.
Media: Chaos! People are in Panic!
What really happened: We just took blankets, beer and had a great time all together.
mag retweeted
Grok 3 is gonna answer whether you can speed run into becoming a frontier lab if you had super large cluster (100k GPUs), insane funding (10B+) and one of the most dedicated teams in the world. Rooting for xAI’s release
It's 11:30pm, and many @xai people are in office, hard working at their computers. It's an amazing vibe. Everyone pushes their way to deliver the best experience to you users. Everyone supports everyone. No one fucks no one with politics. You can just do things.
mag retweeted
Replying to @AravSrinivas
There is something magical for the end user when it can see the model "thinking" it does not have anything to do with accuracy just we want to know the process not only the answer.
mag retweeted
There's never been a better time to start learning tinygrad. We are looking to fund an addictive gamified tutorial, but until then there's some great docs here. github.com/mesozoic-egg/tiny…
mag retweeted
I think this is what Ilya saw.
RL just works. No SFT, no fancy trick.
Give the model some feedback, and reasoning emerges.
Wild stuff.
mag retweeted
Replying to @JaredKubin
It's all about writing good code, figuring out how to do FP8 training, figuring out how to overlap communication and computation on GPUs, also having a big cluster in the same DC with infiniband, it won't work on PCIe cards.
mag retweeted
"DeepSeek-R1-Distill-Qwen-1.5B outperforms GPT-4o and Claude-3.5-Sonnet on math benchmarks with 28.9% on AIME and 83.9% on MATH."
1.5B did WHAT?
mag retweeted
Welcome to the party of Qwen2.5 foundation models! This time, we have the biggest release ever in the history of Qwen. In brief, we have:
Blog: qwenlm.github.io/blog/qwen2.…
Blog (LLM): qwenlm.github.io/blog/qwen2.…
Blog (Coder): qwenlm.github.io/blog/qwen2.…
Blog (Math): qwenlm.github.io/blog/qwen2.…
HF Collection: huggingface.co/collections/Q…
ModelScope: modelscope.cn/organization/q…
HF Demo: huggingface.co/spaces/Qwen/Q…
* Qwen2.5: 0.5B, 1.5B, 3B, 7B, 14B, 32B, and 72B
* Qwen2.5-Coder: 1.5B, 7B, and 32B on the way
* Qwen2.5-Math: 1.5B, 7B, and 72B.
All our open-source models, except for the 3B and 72B variants, are licensed under Apache 2.0. You can find the license files in the respective Hugging Face repositories. Furthermore, we have also open-sourced the **Qwen2-VL-72B**, which features performance enhancements compared to last month's release.
As usual, we not only opensource the bf16 checkpoints but we also provide quantized model checkpoints, e.g, GPTQ, AWQ, and GGUF, and thus this time we have a total of over 100 model variants!
Notably, our flagship opensource LLM, Qwen2.5-72B-Instruct, achieves competitive performance against the proprietary models and outcompetes most opensource models in a number of benchmark evaluations!
We heard your voice about your need of the welcomed 14B and 32B models and so we bring them to you. These two models even demonstrate competitive or superior performance against the predecessor Qwen2-72B-Instruct!
SLM we care as well! The compact 3B model has grasped a wide range of knowledge and now is able to achive 68 on MMLU, beating Qwen1.5-14B!
Besides the general language models, we still focus on upgrading our expert models. Still remmeber CodeQwen1.5 and wait for CodeQwen2? This time we have new models called Qwen2.5-Coder with two variants of 1.5B and 7B parameters. Both demonstrate very competitive performance against much larger code LLMs or general LLMs!
Last month we released our first math model Qwen2-Math, and this time we have built Qwen2.5-Math on the base language models of Qwen2.5 and continued our research in reasoning, including CoT, and Tool Integrated Reasoning. What's more, this model now supports both English and Chinese! Qwen2.5-Math is way much better than Qwen2-Math and it might be your best choice of math LLM!
Lastly, if you are satisfied with our Qwen2-VL-72B but find it hard to use, now you got no worries! It is OPENSOURCED!
Prepare to start a journey of innovation with our lineup of models! We hope you enjoy them!
mag retweeted
Llama3.1 405B Base BF16 is available at @hyperbolic_labs! We still offer the FP8 version for now.
Look at this! I just summoned `jabberwacky`
"I don't care that you don't care that I don't care ...." 😂
>> It's turtles all the way down 🐢🐢🐢
>> tab tab tab
now I'm seriously thinking:
Should we @hyperbolic_labs also host Llama 3.1 405B BASE model in BF16? 🤔
mag retweeted
I can't believe it! The BBC reported at a peaceful pro-British protest where 250-300 Muslims turned up and charged at the police chanting 'Allahu Akbar' which enflamed tensions.
Two-tier Keir's "far-right" narrative smashed by the BBC. 👀
mag retweeted
Wow what a shocker the guy with a PhD economics understands the laws of supply and demand.
Friendly reminder that opposing economic growth is morally evil and fundamentally against nature.
mag retweeted
Current AI is penalised for unconventional thinking. Disagreeing with the "expected" take is seen as a bug, increasing loss. RLHF further erases any uncommon insights found in the corpus.
This ruins the point of intelligence: novel problem solving. It’s impotent intelligence.
mag retweeted
artvis is now available on iOS.
Simply upload a photo of your space, and our AI will generate personalized style suggestions, matched from over 400,000 curated artworks available for purchase. Find the perfect piece for your unique interior in seconds!
mag retweeted
I'm proud to announce @artvis_, our Al-powered art&design advisor is now available on the ios app store!
Simply upload a photo of your space, and our Al will generate personalized style suggestions, matched from over 400,000 curated artworks available for purchase. Find the perfect piece for your unique interior in seconds!
I'd love to hear your feedback - DM's are open!


























