An instruction-tuned Danish-English foundation model based on HRM-Text, trained exclusively on permissible post-training data and designed for strong Danish, English, mathematics, code, and reasoning performance at compact scale.
View modelThe complete, growing collection of Danish Foundation Models releases on Hugging Face, spanning text-generation and multimodal models, Danish-focused continued pre-training, post-training experiments, and deployable model formats.
View collectionA supervised fine-tune of Qwen3.5-27B for psychologically informed refusals, providing structured, supportive responses to high-risk requests while declining harmful guidance and directing users toward appropriate professional resources.
View modelA mixture-of-experts Qwen3.5 model fine-tuned for psychologically informed refusals, designed to replace blunt non-compliance with supportive, harm-preventive communication and relevant professional-resource referrals.
View modelThe first Munin release from Danish Foundation Models: Qwen3.5, Apertus, and Ministral 3 base models post-trained for Danish and English language use.
View collectionA collection of 80 LoRA adapters across four instruct models, each concealing one of 20 secret words while reliably hinting at it, designed as interpretability targets with validated behavior.
View collection