Casper Hansen Profile
Casper Hansen

@casper_hansen_

Followers
8K
Following
2K
Media
393
Statuses
3K

NLP Scientist | AutoAWQ Creator | Open-Source Contributor

Joined August 2019
Don't wanna be here? Send us removal request.
@casper_hansen_
Casper Hansen
3 months
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 🥳
Tweet media one
8
11
140
@casper_hansen_
Casper Hansen
6 hours
0
0
1
@casper_hansen_
Casper Hansen
6 hours
AutoTriton: Automatic Triton Programming with Reinforcement Learning in LLMs. -> map PyTorch code to Triton kernels.-> beats DeepSeek R1 by using SFT+RL on Qwen3 8B.-> combines rule-based and execution-based rewards, avoiding reward hacking, e.g. by a simple is_Triton() reward
Tweet media one
1
0
11
@casper_hansen_
Casper Hansen
7 hours
RL was only a small portion of o3 compared to pretraining. Grok 4 is supposedly 50% split. How will this change the world other than \boxed{}?.
1
0
4
@casper_hansen_
Casper Hansen
8 hours
I remember ICML 2024 where everyone ganged up on Albert Gu for State Space Models and there was literally no escape because the crowd was so big. Today we see so many new models released as hybrids, so congrats to the authors on industry impact.
0
0
1
@casper_hansen_
Casper Hansen
8 hours
Grok 4 is impressive. BUT. Hill climbing on humanity’s last exam and hitting a plateau at 40-50% indicates that we still have a long way to go before we reach PhD-level intelligence.
0
0
4
@casper_hansen_
Casper Hansen
9 hours
“You are a skilled little expert” is very on-brand for DeepSeek.
0
0
2
@casper_hansen_
Casper Hansen
9 hours
When Grok 4 is able to get 100% on AIME25, I think it’s about time we stop releasing Qwen3 + math papers.
0
0
6
@casper_hansen_
Casper Hansen
9 hours
OpenAI is dropping their open-source model the one week of the year where I’m away on vacation to touch grass without laptop access👌😂.
@Yuchenj_UW
Yuchen Jin
23 hours
The best open-source reasoning model will be dropped next Thursday if everything goes well. OpenAI hasn't open-sourced an LLM since GPT-2 in 2019, so I'm excited. We’re hosting it on Hyperbolic. Buckle up.
Tweet media one
1
0
2
@casper_hansen_
Casper Hansen
21 hours
0
0
1
@casper_hansen_
Casper Hansen
21 hours
Step 2 of many: Last week, I released a biomedical dataset of 521k samples. This week, I released full-text embeddings (32k) with 2048 dimension from Qwen3 4B embedding model.
Tweet media one
1
5
25
@casper_hansen_
Casper Hansen
21 hours
BREAKING: Open-source version of o3-mini dropping next week, according to some random news source familiar with the matter.
Tweet media one
2
0
6
@casper_hansen_
Casper Hansen
1 day
How is this budget for these tasks?. I think *maybe* training could be cheaper and *maybe* even synthetic QA dependent on how well the clustering goes and scale of dataset. 1. ($10) Create initial full-text research articles dataset and publish it.2. ($100) Create embeddings.
0
0
0
@casper_hansen_
Casper Hansen
1 day
This is all literally all you need to train a judge
Tweet media one
2
7
169
@casper_hansen_
Casper Hansen
1 day
While everyone is cooking OpenAI on the timeline just because Zuck stole <10/600 researchers, they are cooking GPT-5 to reset your emotions.
0
0
4
@casper_hansen_
Casper Hansen
1 day
New open-source model coming soon you can create slides agentically and offline.
@Zai_org
Z.ai
1 day
We're excited to announce our latest model in the making: GLM-Experimental, now on Chat!. Though still in the early phases of its training cycle, it's showing powerful frontend coding and agentic capabilities. We've combined these to create an end-to-end
Tweet media one
0
0
1
@casper_hansen_
Casper Hansen
1 day
I’m going to try Claude Code soon. What’s a good default rule set that makes sure the model produces elegant code?.
5
0
3
@casper_hansen_
Casper Hansen
1 day
0
0
2
@casper_hansen_
Casper Hansen
1 day
RLVR for sycophantic models is here. The Hunyuan team at Tencent shows you how to get emotionally "sentient" models on par with GPT-4o
Tweet media one
3
7
41
@casper_hansen_
Casper Hansen
1 day
Daily dose of dopamine spike:.uv pip install vllm.
2
2
42
@casper_hansen_
Casper Hansen
1 day
Congratulations to the HF team on this launch! Open science like this is super important and few teams are doing it.
@_lewtun
Lewis Tunstall
2 days
Really excited to share SmolLM3: a strong, smol reasoner!. > SoTA 3B model.> dual mode reasoning (think/no_think).> long context, up to 128k.> multilingual: en, fr, es, de, it, pt.> fully open source (ckpts, data, code, recipes). Details on the
Tweet media one
0
0
4