Stefan Schweter's picture

Stefan Schweter PRO

stefan-it

AI & ML interests

Flair Library šŸ’•, NER & PoS Tagging, LM Pretraining (mostly encoder-only & encoder-decoder), Historical Language Models, German Language Models

Recent Activity

reacted to anakin87's post with šŸ‘ about 8 hours ago
š—œ š˜š—æš—®š—¶š—»š—²š—± š—® š—Ÿš—®š—»š—“š˜‚š—®š—“š—² š— š—¼š—±š—²š—¹ š˜š—¼ š˜€š—°š—µš—²š—±š˜‚š—¹š—² š—²š˜ƒš—²š—»š˜š˜€ š˜„š—¶š˜š—µ š—šš—„š—£š—¢! šŸ‘‘ šŸ—“ļø āœļø Blog post: https://huggingface.co/blog/anakin87/qwen-scheduler-grpo I experimented with GRPO lately. I am fascinated by models learning from prompts and rewards - no example answers needed like in Supervised Fine-Tuning. After the DeepSeek boom, everyone is trying GRPO with GSM8K or the Countdown Game... I wanted a different challenge, like š˜š—²š—®š—°š—µš—¶š—»š—“ š—® š—ŗš—¼š—±š—²š—¹ š˜š—¼ š—°š—æš—²š—®š˜š—² š—® š˜€š—°š—µš—²š—±š˜‚š—¹š—² š—³š—æš—¼š—ŗ š—® š—¹š—¶š˜€š˜ š—¼š—³ š—²š˜ƒš—²š—»š˜š˜€ š—®š—»š—± š—½š—æš—¶š—¼š—æš—¶š˜š—¶š—²š˜€. Choosing an original problem forced me to: šŸ¤” Think about the problem setting 🧬 Generate data šŸ¤ Choose the right base model šŸ† Design reward functions (and experiencing reward hacking) šŸ”„ Run multiple rounds of training, hoping that my model would learn something. A fun and rewarding šŸ˜„ experience. I learned a lot of things, that I want to share with you. šŸ‘‡ āœļø Blog post: https://huggingface.co/blog/anakin87/qwen-scheduler-grpo šŸ’» Code: https://github.com/anakin87/qwen-scheduler-grpo šŸ¤— Hugging Face collection (dataset and model): https://huggingface.co/collections/anakin87/qwen-scheduler-grpo-680bcc583e817390525a8837
View all activity

Organizations

Bayerische Staatsbibliothek's profile picture flair's profile picture Flax Community's profile picture dumitrescustefan-org's profile picture GermanT5's profile picture BigScience: LMs for Historical Texts's profile picture Universal NER's profile picture BigLAM: BigScience Libraries, Archives and Museums's profile picture Libre Euro Lingua-Alliance's profile picture Lang UK's profile picture BabyLM Challenge's profile picture hmByT5 Preliminary's profile picture hmByT5's profile picture Blog-explorers's profile picture German Wikipedia LMs's profile picture hmBERT's profile picture hmTEAMS's profile picture HIPE's profile picture hmBERT Tiny's profile picture hmBERT 64k's profile picture LSV @ Saarland University's profile picture GERMATRON's profile picture PleIAs's profile picture German LLM Tokenizers's profile picture Occiglot's profile picture Social Post Explorers's profile picture GERTuraX's profile picture Stefmal's profile picture Hugging Face Discord Community's profile picture ScaDS.AI German LLM's profile picture ENGEBA's profile picture Nerdy Face's profile picture TensorFlow Model Garden LMs's profile picture

Posts 5

view post
Post
2300
Wohoo 🄳 I have finished my 2025 GPU workstation build and I am very excited to train new awesome open source models on it.

I built my last GPU workstation 5 years ago featuring an AMD Ryzen 5900X, 64GB of G.SKILL Trident Z RGB on an ASRock X570 Taichi cooled by an Alphacool EisbƤr 420. GPU was a Zotac RTX 3090 AMP Extreme. Unfortunately, I was never satisfied with the case - some Fractal Define 7, as it is definitely too small, airflow is not optimal as I had to open the front door all the time and it also arrived with a partly damaged side panel.

For my new build, I've used the following components: an outstanding new AMD Ryzen 9950X3D with 64GB of Corsair Dominator Titanium (what a name). As a huge Noctua fan - warm greetings to my Austrian neighbors - I am using the brand new Noctua NH-D15 G2 on an ASRock X870E Taichi in an amazing Lian Li LANCOOL III chassis. One joke that only NVIDIA Blackwell users will understand: you definitely need a tempered glass panel to check if your GPU cables/connectors start melting šŸ˜‚ And the best is yet to come: I returned my previously bought Zotac RTX 5090 Solid to the eBay seller (because of... missing ROPs, only NVIDIA Blackwell users will again understand) and bought a Zotac 5090 AMP Extreme INFINITY (yes, the long name indicates that this is the flagship model from Zotac) from a more trustworthy source (NBB in Germany).

I am so happy to start training and fine-tuning new open source models - stay tuned!!!

Articles 1

Article

Fine-tune Flair Models on NER Dataset with šŸ¤— AutoTrain SpaceRunner