MatterGPT and SLICES v2.0.0

SLICES 2.0.0 Changelog

1. Introduced MatterGPT model, a GPT2-based large language model for solid-state materials, specifically designed for inverse design of solid-state materials. It excels in multi-property inverse design. For details, see [ paper ]. Added three new Jupyter tutorials (2.1, 2.2, 2.3) to help users get started with experiments. Add an online demo for MatterGPT [Online Demo for MatterGPT].
2. Updated SLI2Cry algorithm: Replaced M3GNet IAP with CHGNet IAP in the third step, further improving invertibility and enhancing SLICES crystal decoding capabilities. Add an online demo for SLICES [Online Demo for SLICES], allowing online conversion between SLICES and CIF.
3. Set CrystalNN as the default crystal chemical bond recognition algorithm for SLICES encoding.
4. Adopted SLICES strategy=4 as the new default encoding method. This encoding can shorten SLICES length by about 40%, effectively reducing computation for large models. Example of the new encoding:
   C C 0 1 ooo 0 1 +oo 0 1 o+o 0 1 oo+
   This simplifies the offset vector from three separate tokens to one token, streamlining the SLICES string.
5. Relocated new version code files to the 'slices' folder. Changed the SLICES library import command from:
   from invcryrep.invcryrep import InvCryRep
   to the more intuitive:
   from slices.core import SLICES
6. Docker image now uses 'python setup.py develop' to install code from the /crystal/slices folder, facilitating SLICES algorithm development. Users can directly modify source code in /crystal/slices, ensuring the docker-called SLICES library is always up-to-date without reinstallation or environment reconfiguration.
7. The Custodian package is utilized for automating DFT verification calculations with VASP, providing error handling and workflow management capabilities.

SLICES 2.0.0 更新日志

1. 引入MatterGPT模型，这是一个基于GPT2的新材料大语言模型，专门为凝聚态材料逆向设计打造，擅长进行多性质逆向设计。详情请参阅 [ 论文 ]。新增三个Jupyter教程（2.1、2.2、2.3），方便用户快速上手逆向设计新材料。 添加了一个在线演示 [材料大模型在线演示].
2. 更新SLI2Cry算法：将第三步的M3GNet IAP替换为CHGNet IAP，进一步提高可逆性，增强SLICES晶体解码能力。添加了一个在线演示 [SLICES编解码在线演示]. 可以在线对SLICES和CIF进行转换以及数据增广。
3. 将CrystalNN设置为SLICES编码使用的默认晶体化学键识别算法。
4. 采用SLICES strategy=4作为新的默认编码方法。这种编码可以缩短SLICES长度约40%，有效降低大模型的计算量。新型编码示例：
   C C 0 1 ooo 0 1 +oo 0 1 o+o 0 1 oo+
   这种方法将偏移矢量从三个独立的token简化为一个token，有效精简SLICES字符串。
5. 将新版本的代码文件放在slices文件夹中。调用SLICES库的命令从：
   from invcryrep.invcryrep import InvCryRep
   变为更加直观的：
   from slices.core import SLICES
6. Docker镜像现在使用'python setup.py develop'安装/crystal/slices文件夹中的代码，方便进行SLICES算法的开发。用户可以直接在/crystal/slices中修改源代码，确保docker调用的SLICES库始终是最新版本，无需重新安装或配置环境。
7. DFT计算使用custodian进行自动化纠错，提高计算成功率。