DANTINOX: A Unified Framework for Multi-Paradigm Language Modeling

DANTINOX：用于多范式语言建模的统一框架

Abstract: Language generation research increasingly spans three paradigms: autoregressive decoding, discrete masked diffusion, and continuous flow-matching. Comparing them is difficult because each lives in a separate codebase, so measured differences often reflect implementation details rather than the paradigms themselves.

摘要： 语言生成研究正日益涵盖三种范式：自回归解码（autoregressive decoding）、离散掩码扩散（discrete masked diffusion）以及连续流匹配（continuous flow-matching）。由于每种范式都存在于各自独立的代码库中，因此对其进行比较十分困难；所测得的差异往往反映的是实现细节，而非范式本身的特性。

We present DantinoX, an open-source JAX/Flax library in which a single modular Transformer backbone serves all three paradigms. Switching the generation paradigm, attention mechanism, or hardware topology requires only a configuration change, while the backbone architecture, tokenizer, initialization strategy, and training infrastructure remain consistent.

我们推出了 DantinoX，这是一个开源的 JAX/Flax 库，其中单一的模块化 Transformer 主干网络可支持上述所有三种范式。切换生成范式、注意力机制或硬件拓扑结构仅需更改配置，而主干架构、分词器（tokenizer）、初始化策略和训练基础设施则保持一致。

This enables controlled cross-paradigm comparisons within one API for training, streaming inference, and benchmarking.

这使得在单一 API 内进行受控的跨范式比较成为可能，涵盖了训练、流式推理和基准测试等环节。

---

Paper Details:

• Authors: Marco Simoni, Aleksandar Fontana, Giulio Rossolini, Andrea Saracino
• arXiv ID: 2609.17535
• Subject: Computation and Language (cs.CL)
• Submission Date: 11 Jul 2026

论文详情：

• 作者： Marco Simoni, Aleksandar Fontana, Giulio Rossolini, Andrea Saracino
• arXiv ID： 2609.17535
• 学科： 计算与语言 (cs.CL)
• 提交日期： 2026年7月11日