Casper Hansen (@casper_hansen

Casper Hansen

5,088 posts

Casper Hansen

@casper_hansen_

ai engineer | previously created autoawq

github.com/casper-hansen

Joined August 2019

Pinned
Casper Hansen
@casper_hansen_
Apr 23, 2025
2.1k stars, 2+ million downloads, and 7000+ models on Huggingface later, and I am officially ready to retire my long-time project AutoAWQ ⚡️ Proud to say that AutoAWQ has been adopted by the @vllm_project and will now be maintained by 55+ contributors 🥳
48K
Casper Hansen
@casper_hansen_
Jan 23, 2025
Am I hallucinating or is DeepSeek with Web Search just better than ChatGPT/Perplexity? Their chat product is really strong in my opinion, and it's completely free.
610K
Casper Hansen
@casper_hansen_
Jan 23, 2025
DeepSeek R1 in Humanity's Last Exam beats all other models on text subset. This is just confirming what we already know - R1 is a real gem.
407K
Casper Hansen
@casper_hansen_
Jun 29, 2025
>be microsoft >acquire wizardlm team >do absolutely nothing and riddle them with politics >wizardlm team leaves for tencent in may >wizardlm team delivers SOTA reasoning model in june biggest microsoft fumble ever
223K
Casper Hansen
@casper_hansen_
Jul 19, 2024
Apple released a 7B model that beats Mistral 7B - but the kicker is that they fully open sourced everything, also the pretraining dataset 🤯
apple/DCLM-7B · Hugging Face
From huggingface.co
229K
Casper Hansen
@casper_hansen_
Mar 1, 2025
Deepseek has an 84.5% profit margin on their pricing. This is better margins than NVIDIA which is crazy to me
223K
Casper Hansen
@casper_hansen_
May 30, 2025
Mysterious package received 😅
160K
Casper Hansen
@casper_hansen_
Jan 20, 2025
The DeepSeek R1 training procedure confused me at first. My brain refused to accept this powerful model could be incredibly straightforward. Let me break down this elegant beast for you 🧵
269K
Casper Hansen
@casper_hansen_
Jan 27, 2025
DeepSeek open-sources Janus Pro, beating Stable Diffusion and OpenAI's DALL-E 3🤯
220K
Casper Hansen
@casper_hansen_
Feb 24, 2025
Anthropic just dropped Claude Code—a real terminal app, no fluff with 70% performance on SWE Bench. No steep learning curve, unlike Aider.
00:00
310K
Casper Hansen
@casper_hansen_
Aug 2, 2025
Researchers designed a 27M model inspired by the brain. It outperforms o3-mini, R1, and others. This makes every single AI lab look silly! Here are their 5 core techniques: - Brain-inspired hierarchical processing, temporal separation, and recurrent connectivity - Hierarchical
177K
Casper Hansen
@casper_hansen_
Jan 16, 2025
I cancelled my Perplexity Pro subscription. I just can't justify it any longer.
457K
Casper Hansen
@casper_hansen_
Dec 25, 2024
I hate to be that guy, but you can't retroactively update the license. There will now forever be an Apache 2.0 licensed version of QvQ that you can git checkout
275K
Casper Hansen
@casper_hansen_
Sep 21, 2025
imagine beating gemini 2.5 pro at OCR with 3B parameters
144K