划分小节吗？先看内容结构：首先讲DLM和AR模型的不同，然后提出BYOD方法，接着讲转换的模型，参数占比，资源需求，效果评估，限制，最后提供资源和结论。首先，标题部分？原文开头是介绍背景，然后方法，所以可能用
把关键部分分一下？不对，先看用户要求，长文用
/
划分小节。首先，原文可以分成几个部分：1. 背景与方法概述；2. 转换模型与资源消耗；3. 效果评估与限制；4. 资源提供与结论。然后每个部分用
包裹？或者直接按自然段落？不对，原文是一个长段落，但里面有不同的内容块。首先，第一句讲AR和DLM的不同，然后讲BYOD方法，然后讲转换的模型，然后讲参数和训练资源，然后讲生成效果，然后讲限制，然后讲提供的资源，最后结论。那首先，用
来划分小节，比如：
Background and Proposed Method
然后对应的，然后
Conversion Results and Resource Consumption
，然后，然后
Generation Performance and Limitations
，然后，然后
Released Resources and Conclusion
，然后？等一下，要保留原文所有信息，不得增删改。那原文的结构：首先，第一部分：“Unlike autoregressive (AR) models, diffusion language models (DLMs) permit parallel token updates, flexible generation, and bidirectional attention, but leading systems rely on costly pretraining or full-model fine-tuning. We present Build Your Own DLM (BYOD), a generalizable LoRA-only recipe for parameter-efficient AR-to-DLM conversion.” 这部分是背景和方法，所以可以用
Context and Method
，然后包裹这部分？然后下一部分：“Without architecture-specific tuning, BYOD induces diffusion generation in Gemma 2 9B, Llama 3.1 8B, Qwen2.5 7B, and Ministral 8B; to our knowledge, this is the first reported AR-to-DLM conversion of a Mistral model. The adapters comprise 4.1-5.8% of final parameters while every base-model parameter remains frozen; each 25,000-iteration conversion needs at most 102.4M tokens, processed in 3.1-4.0 hours on one Google Colab GPU.” 这部分是转换的模型和资源，所以
Conversion Targets and Resource Usage
，然后。然后下一部分：“All four models generate from fully masked sequences, use bidirectional context in a controlled attention test, and retain substantial parent-model knowledge. BYOD-Gemma is the strongest BYOD conversion on our nine-task evaluation, reaching 50.6% average accuracy compared with 51.9% for LLaDA 8B Instruct in the same harness.” 这是生成效果，然后“Open-generation quality at low network function evaluation (NFE) budgets remains the principal limitation: BYOD perplexity is 8.3-11.1 versus 4.1 for LLaDA and 2.2-3.0 for the AR parents when generating 128 tokens in 64 steps.” 这是限制，所以可以用
Performance Evaluation and Limitations
，然后包含效果和限制？然后最后部分：“To enable further experimentation and inspection, we provide a standalone training and evaluation implementation and interactive demos for all four conversions. BYOD thus provides a practical route for extending diffusion generation to other existing model families while identifying the gap that efficient conversion must next close.” 这是资源和结论，用
Released Resources and Conclusion
，然后。现在，检查每个部分：然后，重要的句子比如“BYOD induces diffusion generation in Gemma 2 9B, Llama 3.1 8B, Qwen2.5 7B, and Ministral 8B; to our knowledge, this is the first reported AR-to-DLM conversion of a Mistral model.” 可以用吗？比如Without architecture-specific tuning, BYOD induces diffusion generation in Gemma 2 9B, Llama 3.1 8B, Qwen2.5 7B, and Ministral 8B; to our knowledge, this is the first reported AR-to-DLM conversion of a Mistral model.？或者里面的“first reported AR-to-DLM conversion of a Mistral model”用？比如：“Without architecture-specific tuning, BYOD induces diffusion generation in Gemma 2 9B, Llama 3.1 8B, Qwen2.5 7B, and Ministral 8B; to our knowledge, this is the first reported AR-to-DLM conversion of a Mistral model.” 然后，“BYOD-Gemma is the strongest BYOD conversion on our nine-task evaluation, reaching 50.6% average accuracy compared with 51.9% for LLaDA 8B Instruct in the same harness.” 可以用BYOD-Gemma is the strongest BYOD conversion吗？或者直接保留。现在，开始排版：首先，
Background and Method
Unlike autoregressive (AR) models, diffusion language models (DLMs) permit parallel token updates, flexible generation, and bidirectional attention, but leading systems rely on costly pretraining or full-model fine-tuning. We present Build Your Own DLM (BYOD), a generalizable LoRA-only recipe for parameter-efficient AR-to-DLM conversion.
Conversion Targets and Resource Consumption
Without architecture-specific tuning, BYOD induces diffusion generation in Gemma 2 9B, Llama 3.1 8B, Qwen2.5 7B, and Ministral 8B; to our knowledge, this is the first reported AR-to-DLM conversion of a Mistral model. The adapters comprise 4.1-5.8% of final parameters while every base-model parameter remains frozen; each 25,000-iteration conversion needs at most 102.4M tokens, processed in 3.1-4.0 hours on one Google Colab GPU.
Generation Performance and Evaluation
All four models generate from fully masked sequences, use bidirectional context in a controlled attention test, and retain substantial parent-model knowledge. BYOD-Gemma is the strongest BYOD conversion on our nine-task evaluation, reaching 50.6% average accuracy compared with 51.9% for LLaDA 8B Instruct in the same harness.
Principal Limitation
Open-generation quality at low network function evaluation (NFE) budgets remains the principal limitation: BYOD perplexity is 8.3-11.1 versus 4.1 for LLaDA and 2.2-3.0 for the AR parents when generating 128 tokens in 64 steps.
Released Resources and Conclusion
To enable further experimentation and inspection, we provide a standalone training and evaluation implementation and interactive demos for all four conversions. BYOD thus provides a practical route for extending diffusion generation to other existing model families while identifying the gap that efficient conversion must next close.
这样对吗？检查规则：1. 只输出HTML片段，没有Markdown围栏，没有解释。2. 每个自然段落用包裹，长文用