DFM Mimir v1：一款僅使用許可後訓練數據、以10億參數達成前沿性能的開源HRM

摘要

當前大型語言模型的開發依賴於規模龐大、且往往未經正式授權的資料集，這對致力於開源與符合倫理來源資料的研究者構成了極高的門檻。我們推出 Mimir v1，這是一個基於階層式推理模型（Hierarchical Reasoning Model, HRM）架構的十億參數語言模型，從零開始訓練，僅使用合規的後訓練資料，即在英語方面展現出極具競爭力的表現，並在丹麥語方面樹立了新的最先進水準。Mimir v1 使用 161 個資料集的混合進行訓練，在英語、數學與程式碼及丹麥語的 20 項基準測試中，超越了原始的 HRM-Text 1B，並可與 Qwen 3.5 4B 和 Gemma 4 E2B 等較大型的前沿模型相抗衡。該模型已於 Hugging Face Hub 上公開：https://huggingface.co/danish-foundation-models/DFM-Mimir

English

Current large language model development relies on massive, often non-permissible datasets, creating a high barrier for researchers committed to open-source and ethically sourced data. We introduce Mimir v1, a 1-billion-parameter language model based on the Hierarchical Reasoning Model (HRM) architecture, that is trained from scratch and delivers highly competitive performance for English and sets a new state of the art for Danish using only permissible post-training data. Trained on a mixture of 161 datasets, Mimir v1 outperforms the original HRM-Text 1B and competes with larger frontier models like Qwen 3.5 4B and Gemma 4 E2B, tested across 20 benchmarks for English, Math & Code and Danish. The model is available on the Hugging Face Hub: https://huggingface.co/danish-foundation-models/DFM-Mimir